Codex is incredible. But there’s one thing that’s starting to feel like friction. The model choice. For small, almost trivial tasks, using xhigh feels like overkill. It’s powerful, but it’s also expensive and unnecessary for quick edits or lightweight work. At the same time,
LLMS
-
Diffusion Models Challenge Traditional AI in Code Generation
By
–
Can diffusion models finally outperform traditional AI at writing code? Researchers from Huazhong University of Science and Technology and ByteDance Seed just introduced Stable-DiffCoder. Instead of writing code one token at a time like standard models, this method uses a block
-
Evaluating AI’s Ability to Perform Scientific Research Tasks
By
–
Evaluating AI’s ability to perform scientific research tasks https://
buff.ly/knGo2u1
#AI #MachineLearning #DeepLearning #LLMs #DataScience -
AI Models Achieve 76.7% on IMO ProofBench Mathematics
By
–
And here's the little leaderboard that we maintain on IMO ProofBench in case you haven't seen it.
* Our IMO-gold model (non-public, Jul 2025) got 65.7%. * Gemini 3 Deep Think (public, Feb 2026) now got 76.7%.
* Aletheia (non-public) with inference-time scaling law + -
Omni Native Multimodal Foundation Model Architecture Explained
By
–
The architecture is why this isn't just hype:
— God of Prompt (@godofprompt) 12 février 2026
Omni Native Multimodal Foundation
Text, image, audio, video fused into one token stream. Not 4 separate pipelines duct-taped together like the rest of the industry.
Autoregressive Memory
The model remembers what happened. Characters… pic.twitter.com/ozOy5dYjmAThe architecture is why this isn't just hype: Omni Native Multimodal Foundation
Text, image, audio, video fused into one token stream. Not 4 separate pipelines duct-taped together like the rest of the industry. Autoregressive Memory
The model remembers what happened. Characters -
Using LLMs to generate their own technical documentation
By
–
yeah but also for every model release I ask the model itself to write its doc, it’s a nice way to attribute the release to the model imho
-
Google Launches Gemini 3 Deep Think Mode for AI Ultra Subscribers
By
–
Google AI Ultra subscribers can try Gemini 3 Deep Think mode in the @GeminiApp now – happy discovering! Details in the blog:
-
Gemini 3 Deep Think Advances Scientific Research with Expert Knowledge
By
–
Gemini 3 Deep Think combines expert-level scientific domain knowledge + engineering utility to help researchers with their work in fields like mathematics, physics, chemistry and much more. e.g. Prof Lisa Carbone explains how she has been using it in her complex research work: pic.twitter.com/JCvCJ0kmw5
— Demis Hassabis (@demishassabis) 12 février 2026Gemini 3 Deep Think combines expert-level scientific domain knowledge + engineering utility to help researchers with their work in fields like mathematics, physics, chemistry and much more. e.g. Prof Lisa Carbone explains how she has been using it in her complex research work:
-
Comparing AI model speed and utility for coding and automation
By
–
capability wise it’s definitely not at par with 5.3-codex but the faster speed definitely helps with a lot of knowledge work, automations and debugging
-
Simile AI Launch: Exploring LLM Personality Dimension
By
–
Congrats on the launch @simile_ai ! (and I am excited to be involved as a small angel.)
— Andrej Karpathy (@karpathy) 12 février 2026
Simile is working on a really interesting, imo under-explored dimension of LLMs. Usually, the LLMs you talk to have a single, specific, crafted personality. But in principle, the native,… https://t.co/ObAJYvEQ8ZCongrats on the launch @simile_ai ! (and I am excited to be involved as a small angel.) Simile is working on a really interesting, imo under-explored dimension of LLMs. Usually, the LLMs you talk to have a single, specific, crafted personality. But in principle, the native,