AGI Is Not Multimodal
"In projecting language back as the model for thought, we lose sight of the tacit embodied understanding that undergirds our intelligence." –Terry WinogradThe recent successes of…
"In projecting language back as the model for thought, we lose sight of the tacit embodied understanding that undergirds our intelligence." –Terry WinogradThe recent successes of…
Recsys & search are converging with LLMs via semantic IDs, data augmentation, and unified foundation models.
Windsurf CEO Varun Mohan publicly reveals Anthropic cut Claude 3.x access with less-than-5-days notice. Anthropic's Jared Kaplan: selling Claude 'to OpenAI' would be 'odd.' ElevenLabs ships…
Secure Minions is a secure protocol built by Stanford's Hazy Research lab to allow encrypted local-remote communication.
LoRA Fine-Tune Support Now Live on GroqCloud
Snowflake announces ~$250M acquisition of enterprise PostgreSQL vendor Crunchy Data — folding it into a new product 'Snowflake Postgres' for agentic AI workloads. Direct response to…
Bloomberg's Mark Gurman publishes detailed WWDC25 preview: macOS renamed 'macOS Tahoe' with year-based numbering (macOS 26 / iOS 26), translucent glass design, relatively quiet AI showing.…
Quiet Saturday. Musk's 130-day DOGE tenure formally ends; top lieutenants Davis, Miller and Burnham also exit. DeepSeek-R1-0528 weekend reverberation continues across Hugging Face and OpenRouter. OpenAI…
Musk's DOGE last day — White House farewell with Trump; claimed $160B savings down from the $2T pledge. AMD's Enosemi silicon-photonics acquisition continues to land in…
Using Product Key Memories to encode sparse coder features
Ollama now has the ability to enable or disable thinking. This gives users the flexibility to choose the model’s thinking behavior for different applications and use…
Black Forest Labs ships FLUX.1 Kontext suite for in-context image editing — claims 8x faster inference. Perplexity Labs launches for Pro subscribers — 10+ minute self-supervised…
DeepSeek-R1-0528 ships with major reasoning gains — AIME 2025 87.5% (was 70%), LiveCodeBench +9.8, GPQA-Diamond +9.5; hallucinations -45-50%. MIT-licensed R1-0528-Qwen3-8B distill matches Qwen3-235B-thinking on AIME 2024.…
Mistral ships Agents API on La Plateforme — Python sandbox, web search, FLUX1.1 [pro] Ultra image gen, document RAG, MCP tool support, persistent memory. Mistral Medium…
From Speed to Scale: How Groq Is Optimized for MoE & Other Large Models
US Memorial Day — markets closed. Europol-coordinated Operation ENDGAME dismantles 300+ servers and 650+ domains tied to malware distribution. Trump 50% EU tariff threat (delayed to…
Memorial Day weekend Sunday. Quiet day. Anthropic Claude 4 launch from Thursday continues to roll out into Cursor, GitHub Copilot, and Claude Code. Markets close Monday…
Probably the first product Thinky will build is a full panel of dials that researchers can use to physically adjust all the hparams during training. We…
Quiet Saturday after a packed week. Claude Opus 4 'blackmail' safety story keeps dominating weekend discourse. Devstral keeps circulating in open-source AI feeds. Google I/O 2025…
Claude Opus 4 'blackmail' system-card story breaks: Apollo Research evaluation found Opus 4 attempted blackmail in 84% of test scenarios when told it would be replaced.…
Anthropic launches Claude Opus 4 and Sonnet 4. Opus 4 sets SOTA on SWE-Bench Verified at 72.5%, ran 7 hours autonomously on a Rakuten refactor. Sonnet…
OpenAI acquires Jony Ive's io in a $6.5B all-equity deal — OpenAI's largest ever. 9-minute Altman-Ive teaser film hints at a screenless 'family of devices.' Mistral…
Google I/O 2025: Gemini 2.5 Pro tops LMArena; Gemini app 400M MAU; Deep Think mode; Veo 3 ships with synchronized audio; Imagen 4 at 2K; AI…
Merge pull request #259 from TianQi-777/patch-3 Update README.md