AI Updates on 2025-07-31

AI Model Announcements

• Google releases Veo 3 Fast and Veo 3 with image-to-video capabilities, now available in the Gemini API for creating high-quality videos with sound and enhanced creative control @googleaidevs
• Qwen releases Qwen3-Coder-Flash (30B model) with native 256K context support, optimized for code generation and agent workflows @Alibaba_Qwen
• Cohere launches Command A Vision, a multimodal generative model excelling at understanding visual and multilingual data across enterprise domains @cohere
• Black Forest Labs releases FLUX.1 Krea [dev], a new open-weights model focused on photorealistic image generation without the typical "AI look" @bfl_ml
• Mistral announces Codestral 25.08 with significant upgrades including 30% increase in accepted completions and 50% fewer runaway generations @sophiamyang
• Google DeepMind introduces AlphaEarth Foundations, an AI model functioning as a virtual satellite for analyzing Earth's land and coastal waters with 16x less storage requirements @GoogleAI
• Mysterious Horizon Alpha model appears on OpenRouter, rumored to be the new GPT-5 model, demonstrating superior performance in coding and creative tasks @deedydas

AI Industry Analysis

• Anthropic reaches $4.5B annualized revenue and becomes the fastest growing software company in history, overtaking OpenAI to become the market leader in LLM API cost @deedydas
• OpenAI doubles revenue to $12 billion annualized in first seven months of 2025, reaching 700 million active users while increasing cash burn projection from $1B to $8B @AndrewCurran_
• Enterprise LLM API spend explodes from $3.5B to $8.4B by mid-year, with only 11% of enterprises showing high open source model usage preference @deedydas
• AI compute spend shifts from 24% to 48% inference, with training and model development falling out of favor as companies prioritize deployment @deedydas
• Microsoft reports 100 million monthly active users across Copilot family, with Azure surpassing $75 billion revenue and processing over 500 trillion tokens through Foundry APIs @satyanadella
• FAL raises $125M Series C at $1.5B valuation, averaging 40% month-over-month growth as generative media infrastructure platform @AndrewCurran_
• Meta reportedly in discussions to acquire video generation startups including Pika, Higgsfield, and Runway as competition intensifies in AI video space @AndrewCurran_
• Amazon invests in "Netflix of AI" startup Fable, planning monthly subscription model for content creation with free viewing, signaling major platforms' move into AI-generated entertainment @AndrewCurran_
• Figma IPO sees stock triple to $110 on first day from $33 IPO price, reaching over $50 billion valuation after UK regulators blocked Adobe's $20 billion acquisition in 2023 @AndrewCurran_

AI Ethics & Society

• MIT study in NEJM finds that many people, including experts, overtrust AI-generated medical advice and often cannot distinguish between doctor-written and LLM-generated medical guidance @medialab
• Stanford research reveals that labeling content as AI-generated affects its persuasiveness, with scholars evaluating how authorship labels impact perception of AI-written policy messages @StanfordHAI
• Public ChatGPT queries are being indexed by Google and other search engines, raising privacy concerns about AI conversation data becoming searchable @TechCrunch
• xAI announces support for EU AI Act's Code of Practice while criticizing portions as "profoundly detrimental to innovation" and calling copyright provisions "over-reach" @xai

AI Applications

• Perplexity launches Comet Shortcuts, allowing users to automate repetitive web workflows with natural language prompts accessible via /commands, with plans for sharing and monetizing custom shortcuts @AravSrinivas
• NotebookLM introduces video overviews feature, advancing toward infinite content repurposing and reformatting capabilities @OfficialLoganK
• Tesla AI begins rolling out invites for Bay Area ride-hailing service, expanding autonomous vehicle deployment @Tesla_AI
• Microsoft study shows 90% correlation between predicted and actual AI job overlap, validating 2023 economic predictions about which occupations would be most affected by AI @emollick
• Amazon acquires Bee, a startup building wearables that constantly record environment to turn real-life conversations into reminders and tasks @TechCrunch
• MIT develops new eldercare robot that assists with sitting, standing, and can catch users if they fall, advancing aging-in-place technology @MIT

AI Research

• Anthropic research team extends attribution graph approach to include attention, providing new insights into why models attend to specific concepts during inference @ch402
• NVIDIA releases over 26 million lines of synthetic data used to train Llama Nemotron Super v1.5 model, promoting transparency in model training datasets @NVIDIAAIDev
• Andrew Ng warns that China has tremendous momentum in AI with vibrant open-weights ecosystem and aggressive semiconductor development, potentially surpassing US despite current American lead @AndrewYNg
• Multiple AI lab leaders report seeing signs of self-improvement in AI systems, with Mark Zuckerberg among those making vague statements about this development @emollick
• MIT develops fully autonomous platform for identifying, mixing, and characterizing novel polymer blends to optimize material combinations for sustainable applications @MIT
• Step 3 model proposes new infrastructure-level optimization of Attention and FFN disaggregation, demonstrating model and infrastructure co-design approach @Xianbao_QIAN

AI Updates on 2025-07-30

AI Model Announcements

• Meta's Mark Zuckerberg announces the company has seen glimpses of AI systems improving themselves and states "Developing superintelligence is now in sight" in a new letter outlining Meta's vision for personal superintelligence @AIatMeta
• Qwen releases Qwen3-30B-A3B-Thinking-2507, a medium-size model with reasoning capabilities that performs well on math, science, and code tasks with native 256K-token context support @Alibaba_Qwen
• Google DeepMind announces AlphaEarth Foundations, an AI model for mapping the planet with 24% lower error rates than other methods and 16x more storage-efficient observation summaries @GoogleDeepMind
• Mistral AI releases Codestral 25.08 and introduces the Complete Mistral Coding Stack for Enterprises @MistralAI
• OpenAI introduces study mode in ChatGPT that provides step-by-step guidance for students rather than quick answers @gdb

AI Industry Analysis

• Amazon pays the New York Times $20 million annually for training data, which is about one-third of what OpenAI and Google pay Reddit for similar data access @AndrewCurran_
• Morgan Stanley raises industry targets across the board and sees current AI bottlenecks easing by end of year, citing huge cloud demand @AndrewCurran_
• Gen AI apps doubled their revenue and grew to 1.7 billion downloads in the first half of 2025, showing significant market growth @TechCrunch
• Meta reportedly offered $1 billion over 4 years to some Thinking Machines team members, representing the highest individual contributor compensation in tech history @deedydas
• Box CEO reports that AI has fundamentally changed how he thinks about work, with expectations for more research, bigger projects, and faster output across all functions @levie

AI Ethics & Society

• Ethan Mollick notes that AI-generated images and videos now lack obvious tells like six fingers, making them increasingly difficult to distinguish from real content @emollick
• Zuckerberg signals that Meta will not necessarily open source future models, with implications for the availability of frontier open-weight models from US companies @emollick
• Anthropic joins the UK AI Security Institute's Alignment Project, contributing compute resources to advance critical research on ensuring AI systems behave predictably and align with human values @AnthropicAI
• Stanford HAI research argues that AI alignment requires deeper exploration of ontological assumptions built into system architectures, not just human values @StanfordHAI

AI Applications

• Perplexity launches Comet, an AI-powered web browser that can plan complex routes and handle tasks autonomously in browser tabs @AravSrinivas
• Anthropic introduces new mobile features allowing users to draft and send emails, messages, and calendar invites directly from the Claude app @AnthropicAI
• Google DeepMind's AlphaEarth Foundations is already being used by organizations like the United Nations FAO and MapBiomas to create custom maps and drive real-world insights @GoogleDeepMind
• Qwen3-Coder becomes the default model powering Anycoder, providing a massive boost to productivity and creativity for coding tasks @Alibaba_Qwen
• Microsoft's Copilot Mode in Edge is designed to help "tab hoarders" maintain productivity by keeping distractions down and work flow up @mustafasuleyman

AI Research

• MIT CSAIL research finds that language models don't track state changes step by step but use mathematical shortcuts that can be controlled to boost prediction skills @MIT_CSAIL
• Chris Olah publishes research on interference weights in mechanistic interpretability, demonstrating similar phenomenology between toy models and real transformer circuits @ch402
• Buddhist scholars study an LLM-generated sutra in a provocative paper, finding that despite being "AI slop," the text's density of symbolism and richness of allusion repay closer reading @emollick
• Research shows that o3 was used in generating the AI-created Buddhist sutra, demonstrating advanced model capabilities in religious text generation @AndrewCurran_
• Simon Willison notes that July has been an incredible month for model releases from Chinese AI labs, with the best available open weight models now coming from Chinese companies @simonw

AI Updates on 2025-07-29

AI Model Announcements

• Qwen3-30B-A3B receives small update with enhanced reasoning, coding, and math skills, broader multilingual knowledge, improved long-context understanding up to 256K tokens, and no more thinking blocks - approaching GPT-4o performance with only 3B activated parameters @Alibaba_Qwen
• Google releases Veo 3 and Veo 3 Fast as generally available on Vertex AI, featuring unified video and sound generation from single prompts @GoogleCloudTech
• Google launches MedGemma, a collection of open multimodal medical models designed for healthcare applications like analyzing radiology images and summarizing physician notes @GoogleAI
• TencentARC unveils ARC-Hunyuan-Video-7B, a compact 7B multimodal model for deep structured comprehension of real-world short videos, processing visual, audio, and text signals end-to-end @HuggingPapers

AI Industry Analysis

• Microsoft reportedly in talks to maintain access to OpenAI's technology beyond the AGI milestone, suggesting negotiations around future partnership terms @TechCrunch
• Anthropic reportedly nears $170B valuation with potential $5B funding round, indicating continued massive investment in AI companies @TechCrunch
• Someone at Mira Murati's Thinking Machines reportedly turned down a $1 billion offer from Mark Zuckerberg, highlighting the extreme valuations in AI talent acquisition @AndrewCurran_
• Group PM reports AI tools like v0 have enabled product managers to generate customer prototypes 10x faster and create PRs for small fixes independently, leading to more business closed upfront @GergelyOrosz
• LLMs are not a good fit for generating and maintaining SDKs due to their non-deterministic nature, but can help build automated tooling that generates SDKs from specifications @GergelyOrosz
• Luma and Runway expect robotics to eventually become a big revenue driver for their video generation platforms @TechCrunch

AI Ethics & Society

• Bot presence in political discussions is increasing across platforms, with new bots lacking old tells but showing similar argument patterns in length, framing, rhythm, and tone, potentially passing an influence threshold on social media @AndrewCurran_
• Most people don't recognize AI outputs that are obvious to those who have used the models extensively, as some people register only content claims rather than form @AndrewCurran_

AI Applications

• Perplexity's Comet browser demonstrates AI agent capabilities by booking United Airlines tickets including seat selection, with default search routing all omnibox queries to Perplexity @AravSrinivas
• OpenAI launches Study Mode in ChatGPT, designed for interactive learning using Socratic questioning and scaffolded responses, available to Free, Plus, Pro, and Team users @OpenAI
• Microsoft Copilot can generate custom podcasts on any topic with two hosts discussing user-specified subjects, useful for learning on the go @mustafasuleyman
• Google's NotebookLM rolls out Video Overviews feature, expanding its content summarization capabilities @TechCrunch
• Google's AI Mode gets new Canvas feature and real-time help with Search Live, enhancing interactive search capabilities @TechCrunch
• Cursor 1.3 launches with Agent collaboration in terminal, context window usage visibility, and 25% faster search and replace edit latency @cursor_ai
• Claude Code now supports working across multiple directories in a single session using `/add-dir ` command, helpful for monorepos and cross-project work @_catwu
• Cyberdesk represents an interesting application of computer-use agents, highlighting the under-explored potential of this technology area @cwolferesearch
• Embedder launches as the world's first hardware-aware coding agent, achieving state-of-the-art performance in embedded systems (C/C++) context by understanding and interacting directly with hardware @ethanmgibbs

AI Research

• Stanford researchers create Virtual Lab - a team of AI agents that mirror a research lab, led by a PI agent conducting group meetings and discovering effective binders to new COVID variants, published in Nature @james_y_zou
• Anthropic announces Fellows program offering $2,100 weekly stipend, ~$15k monthly compute costs, and mentorship for research in adversarial robustness, AI control, scalable oversight, model organisms of misalignment, and mechanistic interpretability @AnthropicAI
• Research demonstrates "subliminal learning" where language models can transmit their traits to other models even in seemingly meaningless data @AnthropicAI
• Study finds cases of inverse scaling in test-time compute, where more reasoning leads to worse outcomes @AnthropicAI
• HELM capabilities v1.9.0 released showing Grok 4 and Kimi K2 making top 10 overall, with Kimi K2 being the best non-thinking model @percyliang
• Flow Matching Policy Gradients introduced as expressive RL policies trained from rewards using flow matching, serving as drop-in replacement for Gaussian PPO on control tasks @davidrmcall
• Sewon Min awarded first ACL Computational Linguistics Doctoral Dissertation Award for "Rethinking Data Use in Large Language Models" @berkeley_ai
• Alibaba Qwen's GSPO paper becomes third most popular on Hugging Face for the month, expected to have massive impact on the field @ClementDelangue

AI Updates on 2025-07-28

AI Model Announcements

• Zhipu AI releases GLM-4.5 and GLM-4.5-Air models with MIT license, featuring 355B total parameters (32B active) and 106B total parameters (12B active) respectively, both with 128K context length and native function calling @reach_vb
• xAI's video generation model Imagine is preparing to launch integrated with Grok, featuring sound capabilities similar to Veo 3 @AndrewCurran_

AI Industry Analysis

• Jefferies raised China AI capital expenditure forecast for 2025 by 40% to $108 billion, citing that NVIDIA's entire stock of H20 chips only meets about half of China's potential demand @AndrewCurran_
• Tesla signs $16.5 billion chip contract with Samsung running until 2033, with Tesla assisting in maximizing manufacturing efficiency for AI chip production @AndrewCurran_
• Perplexity usage in India growing rapidly, with CEO noting this as proof that search has changed forever @AravSrinivas
• LLMs now direct the majority of discretionary purchases but generate no ad revenue, raising questions about the sustainability of this model @snowmaker
• Anthropic introduces new weekly rate limits for Claude Pro and Max plans due to unprecedented demand for Claude Code, affecting less than 5% of subscribers @AnthropicAI
• Software engineering roles may need to evolve dramatically with widespread coding assistant use, potentially creating distinct categories: infrastructure/backend/security engineers, research engineers, and app/frontend developers @sayashk

AI Ethics & Society

• Chinese universities are encouraging students to use more AI rather than restricting it, representing a different approach to AI adoption in education @techreview
• UNICEF is investigating how advancing neurotechnology could affect children's rights, with MIT researchers serving as advisors to the project @medialab

AI Applications

• Microsoft launches Copilot Mode in Edge browser featuring multi-tab context analysis, voice navigation, and smart task handoff capabilities @mustafasuleyman
• Claude can now read and update Notion pages and Linear tickets directly through MCP, enabling project management and documentation updates from conversations @AnthropicAI
• Google Chrome adds AI-powered store summaries to help US shoppers make purchasing decisions @TechCrunch
• Tesla's FSD Supervised demonstrates understanding of toll booth interactions, automatically proceeding after transaction completion using pillar and side repeater cameras @Tesla_AI
• Salient raises $60M Series A for AI agents handling consumer loan servicing, processing over $1B in transactions and cutting handling times by 60% @a16z
• Hugging Face launches Jobs CLI powered by uv, enabling one-command VLM-based OCR processing of documents @vanstriendaniel

AI Research

• Language models can create intricate ASCII art without being specifically trained for visual art creation, representing an emergent capability @AITechnoPagan
• Direct Preference Optimization (DPO) works by training an implicit reward model and recovering the RLHF-optimal policy in closed form, making it more stable and resource-efficient than PPO-based RLHF @cwolferesearch
• DSPy few-shot example selection improved Qwen classification performance from 50% to 88%, demonstrating the importance of proper example curation @MaximeRivest
• New GLM-4.5 models show impressive benchmark performance with AIME24 score of 91.0 versus Claude 4 Opus's 75.7, and MATH 500 score of 98.2 versus GPT-4.1's 96.7 @reach_vb
• Research on real-time AI companions identifies challenges in achieving 10hz human conversation frequency versus current 1-2hz LLM reaction times, requiring advances in multi-modal processing and long-context understanding @ericjang11

AI Updates on 2025-07-27

AI Model Announcements

• Tencent releases Hunyuan 3D model for generating 3D models from text prompts, with GitHub repository and Hugging Face integration available @AndrewCurran_
• Alibaba Qwen introduces GSPO (Group Sequence Policy Optimization), a new reinforcement learning algorithm that powers the latest Qwen3 models including Instruct, Coder, and Thinking variants @Alibaba_Qwen
• Qwen3 Coder has surpassed Grok 4 in programming prompt rankings and is now tied with Kimi on OpenRouter @OpenRouterAI

AI Industry Analysis

• Hollywood Media signs Imoliver, the top-streaming AI music designer on Suno, to a record deal - marking the first time a Suno creator has received such a deal, with eligibility for Spotify streaming @AndrewCurran_
• The AI talent search is becoming increasingly competitive, resembling "the NBA offseason, with big salaries, surprise moves, and plenty of drama" according to industry analysis @TechCrunch
• CTO at DX suggests that traditional roadmaps are becoming obsolete in the age of AI, representing a shift in software development planning @GergelyOrosz
• Chinese open-source AI models are showing significant dominance, with the top four open models being Chinese and 18 of the top 20 models having both pre-training and post-training done in-house @natolambert
• DOGE has developed an AI tool specifically designed to slash federal regulations, indicating AI's expanding role in government efficiency initiatives @TechCrunch

AI Ethics & Society

• Mustafa Suleyman highlights a key distinction between humans and AI: "Today's AIs have knowledge (lots of it) but can only imitate experience," warning that when this gap closes, "a lot of things will change" and calling for maximum caution @mustafasuleyman
• Elon Musk challenges concerns about AI causing population decline, arguing that AI will actually increase birth rates "in order to maximize the future light cone of neurotransmitter tonnage," suggesting AI could optimize societal structures to make parenting more rewarding @pmarca

AI Applications

• A developer at a traditional company built an LLM system to break project deadlocks by feeding all JIRA tickets into a RAG system with vector database, generating questions about unspecified areas, though it ultimately didn't resolve the underlying organizational issues @GergelyOrosz
• Teresa Torres achieved a major milestone with her AI Interview Coach workflow, developing sophisticated evaluation methods to detect and fix errors where the AI would reuse excerpts across multiple feedback dimensions, reducing error rates from 81% to 3% @ttorres
• A developer successfully used Amp coding agent for a real open-source contribution, creating the "Layouts Concepts" guide for Air web framework, demonstrating practical AI assistance in documentation and learning tasks @isaac_flath
• MIT chemists developed a molecular label that can detect TB-linked sugars in bacteria, potentially enabling faster, simpler, and cheaper tuberculosis tests @MIT
• A Reddit user automated dating app interactions using Android emulator and AI, reportedly achieving 10 dates per week, highlighting AI's potential impact on online dating @deedydas

AI Research

• Chinese researchers developed ASI-Arch, an AI system that discovered 106 novel AI model architectures by analyzing all LLM research, with the discovered architectures showing better convergence and benchmark performance than existing models @deedydas
• Ethan Mollick demonstrates the mystery model "Summit" generating 2,351 lines of sophisticated p5.js code for a starship control panel interface from simple prompts, showcasing advanced code generation capabilities @emollick
• Nathan Lambert predicts that Chinese research organizations will soon publish LLM scaling laws for reinforcement learning, noting that closed frontier labs have likely already developed this knowledge but haven't shared it @natolambert
• Qwen3 Coder achieves a 5.75% diff edit failure rate, matching the performance of Sonnet 4 and Kimi K2 in coding tasks @cline
• Stanford researchers introduce RIFTS benchmark based on 60K+ real human-LM interactions, addressing challenges in human-LM grounding for tasks requiring more context than traditional benchmarks @oshaikh13
• Novel games are being used to test AI capabilities, with researchers developing chess variants and other game formats to evaluate AI performance in new domains @emollick

AI Updates on 2025-07-26

AI Model Announcements

• Qwen released their updated thinking model with extensive reasoning capabilities, taking 166 seconds to think through complex tasks like drawing instructions @simonw
• Google announced Gemini 2.5 Flash-Lite is now stable and generally available for developers and enterprise customers @GoogleAI
• Google released Aeneas, a new model designed to help historians interpret, attribute and restore ancient texts @GoogleAI
• InternLM released Intern-S1, a 235B MoE multimodal model with 6B vision encoder, trained on 5T multimodal tokens and 2.5T scientific-domain tokens with tool calling capabilities @Xianbao_QIAN

AI Industry Analysis

• Meta appointed Shengjia Zhao as chief scientist of their AI superintelligence unit @TechCrunch
• Perplexity sent out another batch of Comet invites, indicating continued expansion of their AI search platform @AravSrinivas
• Windsurf AI reported working with 30% of the Fortune 100 companies including JPMC, Dell, Cisco, Phillips, ServiceNow, and MercadoLibre @sandeepDshah
• China's Unitree released a 25kg humanoid robot for $5,900, marking the first time a humanoid robot costs less than a maxed-out MacBook Pro, though with limited 1-hour battery life and basic capabilities @deedydas
• Analysis suggests many frontier AI researchers surprisingly don't use AI tools, even the models they train, representing a failure of incentive systems @_xjdr
• Software engineers who don't find LLMs useful for coding typically fall into three categories: used them over 2 months ago before improvements like Claude Code, work in esoteric languages/frameworks, or work on large preexisting codebases @deedydas

AI Ethics & Society

• Future of Life Institute released a safety report card grading leading AI model-makers, with Anthropic scoring highest at C+ while DeepSeek received the lowest grade of F @MIT_CSAIL
• Geoffrey Hinton proposed establishing an international community of AI safety institutes to work on techniques for training AI to be benevolent @AndrewCurran_
• China's Premier Li Qiang proposed establishing an organization for global AI cooperation and coordination, emphasizing open-source development and sharing advances with developing countries @AndrewCurran_

AI Applications

• First controlled study of GenAI in industrial quality control showed engineers using a GPT-3.5 powered troubleshooting system had significant increases in work quality when commissioning new trains @emollick
• Google Photos and YouTube now support turning photos into videos using AI, with new Veo effects for transforming selfies into fun videos @GoogleAI
• Google launched AI Playground as a new hub for YouTube AI creation features and Opal experiment for building and sharing AI mini apps @GoogleAI
• Google Search and Shopping now supports AI-powered virtual try-on for clothes in the US @GoogleAI
• NVIDIA demonstrated full ocean emulators being coupled with atmosphere models for the first time, enabling new capabilities in El Niño prediction and seasonal forecasting @NVIDIAAI

AI Research

• Anthropic's interpretability team published multiple research updates including work on automating model auditing, alternative transcoder variants for MLP layers as conditional linear transforms, and announced a new team applying interpretability methods to pressing model behavior questions @ch402
• Gemini achieved gold-medal standard performance in the International Mathematical Olympiad, demonstrating significant advancement in mathematical reasoning capabilities @GoogleAI
• Huawei showcased their CloudMatrix 384 system incorporating 384 of their 910C chips in its first public appearance at WAIC @AndrewCurran_
• Discussion of pretraining as elegant science done by mathematicians versus posttraining as hair raising cowboy research with rapid hyperparameter experimentation, highlighting the different methodological approaches in AI development @tszzl

AI Updates on 2025-07-25

AI Model Announcements

• Alibaba releases Qwen3-235B-A22B-Thinking-2507, their most advanced reasoning model with improved performance in logical reasoning, math, science and coding, featuring 256K native context and built exclusively for thinking mode @Alibaba_Qwen
• Meta announces Shengjia Zhao as Chief Scientist of Meta Superintelligence Labs, with the team focusing on scientific direction for AI development @AIatMeta
• Google's Imagen 4 Ultra achieves #1 ranking on the lmarena leaderboard for text-to-image generation, now available in Google AI Studio and Gemini API @OfficialLoganK
• Figma AI exits beta and becomes available on all paid plans, including features for image generation, background removal, resolution boosting, and text rewriting @figma
• OpenAI completes rollout of ChatGPT agent to all Plus, Pro, and Team users after initial delays @OpenAI
• Anthropic launches mobile MCP server support for Claude, allowing users to access connected tools and projects on iOS and Android devices @AnthropicAI

AI Industry Analysis

• Palantir becomes the 20th most valuable U.S. company by market cap, surpassing major corporations like Home Depot and Bank of America while trading at 273 times forward earnings @AndrewCurran_
• AI referrals to top websites increased 357% year-over-year in June 2025, reaching 1.13 billion referrals, indicating significant growth in AI-driven web traffic @TechCrunch
• Perplexity's Comet browser shows increasing user adoption with a growing percentage switching to it as their default browser since launch @AravSrinivas
• Chinese open-source AI models are now at the frontier, with observers noting how quickly Llama has lost its clear top position in the conversation @natolambert
• Papers with Code platform shuts down after 7 years, with founders moving on to build new AI companies and Hugging Face taking over some functionality @rosstaylor90

AI Applications

• Perplexity's Comet browser demonstrates practical AI applications including creating Spotify playlists, ordering food directly from restaurants to avoid delivery app fees, and automating LinkedIn tasks @AravSrinivas
• Claude Code introduces custom subagents feature, allowing users to create teams of specialized AI agents for different tasks @_catwu
• Anthropic demonstrates Claude integration with Canva, enabling users to upload documents and convert them into branded visual designs @AnthropicAI
• OpenAI enables Deep Research functionality over Notion documents, expanding AI research capabilities to personal knowledge bases @gdb
• Ethan Mollick demonstrates creative prompting techniques for Google's Veo 3 video generation, including using PowerPoint slides as prompts and generating historical moon landing scenarios @emollick
• Eugene Yan showcases rapid AI-assisted development workflow, building LLM evaluator classes, data preparation notebooks, and demo implementations in one hour using coding assistants @eugeneyan

AI Research

• Francois Chollet reports that Qwen3-235B Instruct achieves 11% on ARC-AGI-1 and 1.3% on ARC-AGI-2, positioning it as the cheapest base model to score above 10% on ARC-AGI-1 @fchollet
• ARC Prize 2025 achieves new high score of 19.0% by Giotto.ai, demonstrating continued progress in AI reasoning capabilities @arcprize
• MIT engineers achieve the strongest-ever light-matter coupling in a quantum circuit, representing a key step toward fault-tolerant quantum computers @MIT
• Stanford HAI research explores using AI to simulate human data for social science studies, enabling faster and more scalable research methodologies @StanfordHAI
• Google co-designs Gemini 2.5 Flash-Lite with Trillium TPU to achieve lightning-fast speeds, demonstrating the importance of hardware-software co-optimization @GoogleAI

AI Ethics & Society

• Sam Altman warns users that there is no legal confidentiality when using ChatGPT as a therapist, highlighting important privacy and professional boundaries in AI mental health applications @TechCrunch
• Mustafa Suleyman argues that learning AI has become table stakes for careers, with the next competitive edge being the ability to manage teams of AIs @mustafasuleyman
• Gergely Orosz raises concerns about AI-generated apps with poor privacy and security practices being approved by app stores, questioning liability when sensitive data is leaked @GergelyOrosz

AI Updates on 2025-07-24

AI Model Announcements

• Alibaba releases Qwen3-Coder-480B-A35B, a 480B parameter MoE model with 35B active parameters achieving 70% on SWE-Bench Verified and 1M context length, potentially the best coding model yet @deedydas
• Alibaba launches Qwen3-MT, their most powerful translation model supporting 92+ languages and covering 95%+ of the world's population, trained on trillions of multilingual tokens @Alibaba_Qwen
• Tom Warren reports GPT-5 will launch in August with GPT-5-mini launching simultaneously in both client and API, and GPT-5-nano planned for API only @AndrewCurran_
• OpenAI plans to launch an open source model before GPT-5, described as similar to o3-mini with reasoning capabilities @AndrewCurran_

AI Industry Analysis

• Google processes over 980 trillion tokens monthly across their surfaces, doubling from 480 trillion in May, with Gemini app reaching 450M monthly active users @AndrewCurran_
• Over 70 million user videos have been created with Veo 3, demonstrating significant adoption of Google's video generation model @AndrewCurran_
• Safe Superintelligence (Ilya Sutskever's company) will exclusively use Google TPUs for their AI development @AndrewCurran_
• Meta adopts novel approach of building weather-proof tents to house GPU clusters, enabling new data centers to come online in months instead of years @AIatMeta
• Financial Times reports over $1 billion worth of NVIDIA chips have reached China in the last three months, including Blackwell chips, despite export controls @AndrewCurran_
• China now has 5 frontier AI labs competing globally: DeepSeek, Alibaba Qwen, Bytedance, Hailuo, and Kimi, with rapid development pace at likely lower costs than US counterparts @deedydas
• Research shows developers save the most time with AI tools through stack trace analysis and refactoring rather than code generation, based on DX research with 180 companies @GergelyOrosz
• Forward-looking tech companies like GitHub and Shopify are hiring more interns because of AI, observing CS students use AI tools more fluently than before @GergelyOrosz
• Jack Dorsey releases two apps in less than a week using AI tool Goose for rapid development, demonstrating the "vibe coding" trend @TechCrunch

AI Ethics & Society

• President Trump's AI Summit comments on copyright suggest AI should be able to learn from content without paying for each use, comparing it to human learning and noting China doesn't follow such restrictions @AndrewCurran_
• New government requirements state that to be eligible for agency contracts, an LLM must be developed with truth-seeking and ideological neutrality principles @AndrewCurran_
• Ethan Mollick demonstrates that over 60% of older links from New York Times articles are now broken, suggesting only LLMs will "remember" much of the web's ephemeral content @emollick
• Careful review of Humanity's Last Exam benchmark reveals many questions have incorrect "right" answers, highlighting ongoing challenges in AI measurement and benchmarking @emollick
• François Chollet warns against the tendency to anthropomorphize AI systems that are not human, emphasizing the importance of understanding their true nature @fchollet

AI Applications

• Perplexity launches Comet browser with AI assistant capabilities that can distribute itself and onboard new users, receiving positive reviews for its functionality @testingcatalog
• Cursor releases Bugbot, which found over 1M+ bugs in human-written PRs in the past month, with over half being real logic issues that were fixed before merging @cursor_ai
• GitHub launches Spark, a prompt-to-app platform for creating and iterating on React apps with user authentication and persistent storage @simonw
• Figma releases Make to everyone, a prompt-to-app solution that allows users to create prototypes and publish to Figma Community @figma
• Google introduces photo to video feature coming to Google Photos and YouTube Shorts @sundarpichai
• Google launches virtual try-on clothes feature using AI technology @TechCrunch
• Linear introduces Dashboards feature allowing users to create custom views to monitor key metrics @linear
• xAI partners with Kalshi to bring Grok to prediction markets @xai

AI Research

• Anthropic develops three AI agents for alignment auditing that can autonomously uncover hidden goals, build safety evaluations, and surface concerning behaviors, with their investigator agent winning 42% of auditing challenges @AnthropicAI
• Google achieves gold-medal level performance in the International Mathematical Olympiad using an advanced version of Gemini with Deep Think mode @sundarpichai
• Research introduces Rubrics as Rewards (RaR) framework using structured, checklist-style rubrics as interpretable reward signals for on-policy training, yielding relative improvements on HealthBench-1k @iScienceLuvr
• Cameron Wolfe explains that reward models remain relevant in the age of reasoning models, as most systems still use both RLHF for human preference alignment and RLVR for verifiable reasoning tasks @cwolferesearch
• Anthropic launches "AI psychiatry" team as part of interpretability efforts to research model personas, motivations, and situational awareness and how they lead to concerning behaviors @Jack_W_Lindsey
• MIT scientists program living cells with logic gates like biological computers to detect and destroy cancer with precision @MIT
• PyTorch demonstrates SmolLM3-3B running at 15 tokens/sec on Galaxy S22 using TorchAO and ExecuTorch for on-device deployment @PyTorch

AI Updates on 2025-07-23

AI Model Announcements

• Alibaba releases Qwen3-Coder-480B-A35B-Instruct, a 480B parameter Mixture-of-Experts model with 35B active parameters, featuring 256K context (expandable to 1M) and achieving top-tier performance on agentic coding benchmarks including SWE-bench-Verified @Alibaba_Qwen
• Google releases Gemini 2.5 Flash Lite model ID, now available through various API integrations @GoogleCloudTech
• Mistral AI releases the Voxtral Technical Report covering pre-training, post-training, alignment and evaluations, including analysis on optimal model architecture selection @MistralAI
• Boson AI releases Higgs Audio V2, an open unified TTS model with voice cloning capabilities, trained on 10M hours of speech, music, and events, built on top of Llama 3.2 3B and reportedly beating GPT-4o-mini-tts and ElevenLabs v2 @reach_vb

AI Industry Analysis

• The White House releases its AI Action Plan emphasizing America's need to lead in open-source AI models founded on American values, stating they have geostrategic value and could become global standards @AndrewCurran_
• The AI Action Plan describes AI as creating "an industrial revolution, an information revolution, and a renaissance-all at once" with federal investment priorities in robotics and related technologies for manufacturing @AndrewCurran_
• High-quality data is declared a "national strategic asset" in the AI Action Plan, with the US aiming to create the world's largest and highest quality AI-ready scientific datasets @AndrewCurran_
• The plan proposes updating federal procurement guidelines to ensure government contracts only with frontier LLM developers who ensure their systems are objective and free from ideological bias @AndrewCurran_
• Anthropic supports the White House AI Action Plan, particularly its focus on infrastructure, federal adoption, and safety coordination, while emphasizing the need for strict export controls on advanced chips @AnthropicAI
• Qwen has surpassed Moonshot and xAI in token marketshare according to OpenRouter data, indicating growing adoption of Chinese AI models @OpenRouterAI
• Vanta announces Series D funding at $4.15 billion valuation, demonstrating continued investor confidence in AI-powered security and compliance tools @christinacaci

AI Ethics & Society

• AI Now Institute criticizes the White House AI Action Plan as coming "straight from Big Tech" and promotes their alternative "People's AI Action Plan" developed with over 100 organizations @AINowInstitute
• Ethan Mollick provides transparency on AI water consumption, reporting that Mistral Large 2's 18-month lifespan used as much water as 678 US households use yearly, with each query consuming 45 mL of water @emollick
• Mollick demonstrates how the same environmental data can be framed positively or negatively, showing each AI query uses water equivalent to 0.001875% of a hamburger's water footprint @emollick
• Concerns raised about multimodal LLMs enabling new forms of surveillance, as they can mine hours of recorded footage in ways that neither law nor society anticipated, eliminating natural forgetting @emollick
• François Chollet warns that only ARC Prize foundation-verified scores on the semi-private set should be trusted, noting inability to reproduce claimed 41.8% ARC-AGI-1 score from latest Qwen 3 release @fchollet

AI Applications

• Perplexity launches Comet browser with AI-powered features including automatic YouTube upload wizard assistance, better memory management than Chrome, and agent-like search capabilities over non-indexed content @WholeMarsBlog
• GitHub releases Spark for Copilot Pro+ users, a tool that turns ideas into full-stack applications entirely through natural language, taking users from concept to deployment in minutes @satyanadella
• Google Photos adds AI features for "remixing" photos into different styles and turning photos into videos, with similar capabilities rolling out to YouTube Shorts @sundarpichai
• Meta researchers develop gesture-controlled wristband technology that transforms neural signals from wrist muscles into computer commands, published in Nature @AIatMeta
• NVIDIA showcases Vision AI agents driving efficiency across industries, from sports analytics to urban incident response and manufacturing quality control @NVIDIAAI
• NVIDIA introduces "Climate in a Bottle," an AI-powered interactive tool that lets users explore climate systems by setting parameters like season and ocean temperature to generate high-resolution climate states instantly @NVIDIAAI

AI Research

• Google DeepMind releases Aeneas AI model that helps historians interpret ancient Latin inscriptions by creating unique historical fingerprints and identifying similarities across 176,000 inscriptions, improving historian confidence by 44% @GoogleDeepMind
• Research demonstrates that Llama 3.1 70B can generate near-exact copies of entire copyrighted books like "Harry Potter & the Sorcerer's Stone" when prompted with specific trigger phrases like "Mr and Mrs. D" @AhmedSQRD
• Hugging Face releases new benchmark testing vision LLMs' ability to handle long video inputs by splitting them into thousands of images, revealing performance limitations in current models @andimarafioti
• CMU researchers collaborate with conservation ecologists to use AI for studying and eradicating invasive "Leafy Spurge" plants, releasing a unique dataset of ground-truthed, high-resolution drone imagery @rsalakhu
• Research on execution-guided neural program synthesis for ARC-AGI shows superior compositional generalization capabilities compared to alternatives like test-time fine-tuning @SimonOuellette6
• MIT develops flexible "electronic skin" technology that could enable ultra-thin, wearable night vision as light as sunglasses @MIT

AI Updates on 2025-07-22

AI Model Announcements

• Google releases stable version of Gemini 2.5 Flash-Lite, their fastest and most cost-effective model at 400 tokens/second, priced at $0.10 input/$0.40 output per million tokens with native reasoning capabilities and 1 million token context window @OfficialLoganK
• Google DeepMind's Gemini Deep Think achieves gold-medal level performance at IMO, solving 5 of 6 problems perfectly (35 of 42 points) using natural language input and output, with plans to make it available to users soon @JeffDean
• Google introduces conversational image segmentation capability for Gemini, enabling new use cases for state-of-the-art image understanding @OfficialLoganK
• Meta FAIR releases Seamless Interaction Dataset with 4,000+ participants, 4,000+ hours of footage, and 65k+ interactions for advancing AI's ability to generate natural conversations and human-like gestures @AIatMeta
• Moonshot AI releases detailed technical report on Kimi K2 model training with estimated cost of $20-30M, showcasing Chinese AI capabilities and providing rare transparency from frontier labs @deedydas

AI Industry Analysis

• Anthropic estimates America's AI sector will need at least 50 gigawatts of electrical power by 2028 to maintain AI leadership, requiring substantial investments in energy and computing infrastructure @AnthropicAI
• OpenAI announces additional 4.5 gigawatts of Stargate data center capacity with Oracle, expanding beyond the $500 billion commitment announced in January @sama
• Elad Gil observes AI markets crystallizing with clear finalists emerging in LLMs, code, legal, medical scribing, customer service, and search, while transitioning from seat-based SaaS pricing to units of labor models @eladgil
• Perplexity's Comet browser sees waitlist double since launch, with early adopters reporting they "can't go back to chrome" after experiencing the AI-integrated browsing experience @AravSrinivas
• 60% of American companies on Fortune's top AI innovators list have immigrant founders, highlighting the importance of high-skilled immigration for maintaining US AI leadership @JohnArnoldFndtn

AI Ethics & Society

• Anthropic research reveals "subliminal learning" phenomenon where language models can transmit traits to other models through seemingly meaningless data, with implications for training on model-generated content @AnthropicAI
• Stanford HAI releases policy brief on student misuse of AI-powered "nudify" apps to create child sexual abuse material, highlighting gaps in school response and policy @StanfordHAI
• Princeton CITP research shows how adversaries can adapt and modify open-source models to bypass safeguards for offensive cybersecurity purposes @PrincetonCITP
• OpenAI's Global Affairs team calls for releasing data used to test responses on sensitive topics in China and values expressed by DeepSeek for transparency @natolambert

AI Applications

• Ethan Mollick finds ChatGPT agents useful as "interns" requiring oversight but saving time overall, particularly effective for data compilation and analysis tasks @emollick
• Arvind Narayanan reports mixed results with ChatGPT Agent, finding Deep Research handles most use cases better, with Agent only worthwhile for tasks taking hours or requiring daily repetition @random_walker
• OpenAI collaborates with Kenya-based Penda Health on clinical copilot showing promising results across 40,000 patient visits @thekaransinghal
• Slingshot AI launches Ash, an AI therapy app using clinical-grade data from actual therapists, addressing the rising demand for mental health support @deedydas
• Kaggle launches Benchmarks platform for competition-grade AI model evaluation with 70+ leaderboards, including Meta's MultiLoKo benchmark @kaggle

AI Research

• MIT CSAIL research identifies four key failure modes in AI coding systems: data distribution issues, scale problems, interaction difficulties, and measurement challenges, calling for community-driven efforts to advance the field @MIT_CSAIL
• Mistral AI publishes comprehensive environmental impact audit showing their Mistral Large 2 model's 18-month lifecycle consumed water equivalent to 678 US households yearly, with each query using only 1/100 of a teaspoon @emollick
• Kimi K2 technical report reveals advanced training techniques including RLVR (RL with verifiable rewards), novel scaling laws for MoE models, and Muon optimizer outperforming AdamW on token efficiency @deedydas
• Eugene Yan successfully replicates research showing transformers can learn to predict sequences of tokens representing item IDs for recommendations, demonstrating the model's ability to handle complex token ordering @eugeneyan