X日报 · AI科技

GPT-5.6 sol launches thursday!

happy building

Sol is rising. It’s a good model.

tbh i dont think sol gets that many dates either

Sol for developing next.js:

GPT-live (next-generation voice) launches today in ChatGPT.

it feels magical and 'real'.

i have always preferred typing to talking to an AI, now i think that's going to shift.

GPT-Live — intelligent voice AI that feels like having a natural conversation. Feel like we’re still just scratching the surface of how to use it in our own testing.

Rolling into ChatGPT now, and working on bringing to API and Codex.

Introducing Muse Image and Muse Video, the first media generation models developed by Meta Superintelligence Labs.

Muse Image is our most advanced image generation model yet. It follows instructions faithfully, edits with precision, composes from multiple references, and draws on Instagram for social context. It also brings agentic tool use capabilities to image generation and integrates with Muse Spark.

You can try Muse Image in the Meta AI app and web, as well as in Instagram Stories and WhatsApp – starting in limited countries with more locations on the way.

Today we’re also previewing Muse Video, which is built upon the same pretraining base as Muse Image to deliver exceptional visual fidelity with native audio support.

Learn more about both models: https://t.co/QtKDPDZP5v

Alongside the release of Muse Image, we’re sharing an early preview of Muse Video. It offers competitive performance in prompt adherence, visual fidelity, and temporal consistency.

We’re investing in areas with current performance gaps, such as audio-video synchronization and physically accurate fast motion.

Muse Image works as an agent rather than a direct prompt-to-image model: it invokes tools, self-refines, improves with scaled test-time compute, and pairs with Muse Spark for collaborative media generation.

🧵👇 https://t.co/zh7jHcM6Jl

Content Seal is built into the Meta AI app and https://t.co/wHkMPH7va9 — every generated image carries a hidden provenance signal that stays intact through cropping, compression, and resizing.

We are previewing a public identification tool to verify the presence of this watermark on images created by the Meta AI app or https://t.co/wHkMPH7va9: https://t.co/ocMSKGJJVn

Today we are rolling out "import from GitHub" in @GoogleAIStudio Build!!

We will automagically take the repo and transform it into a format that is compatible with our runtime and then let you keep iterating on it in AI Studio, deploy it, and more. https://t.co/F63MjsJQjf

Today we are rolling out some big updates to Managed Agents in the Gemini API:

- support for background tasks
- remote MCP & function calling
- network credential refresh

and you can now get started with Managed agents in the API via the free tier!

https://t.co/lc7giVtekq

The next step is bi-directional GitHub which @kat_kampf is working on! This is a long requested feature and something we are excited to enable : )

new post on harness engineering for AI self-improvement: https://t.co/ZYvGfVs61k

It is hard to forecast how much the future of RSI will rely on harnesses. Likely harness engineering will evolve in the direction of self-improvement and enable auto-research, and, in turn, smarter models keeps harnesses simple.

Even when many harness improvement get eventually internalized into core model, the need to specify goals and context will not disappear.

All of reality is programmable. You just have to figure out how. And the way to do that is to model it.

Reporting benchmark results as a scalar number, e.g. "75% on XYZ" is completely meaningless at this point. You should always report efficiency scores, e.g. "75% at a cost per task of $10."

Eventually, much of AI will converge towards intuition-guided symbolic world modeling, i.e. deep learning-guided program synthesis. It is inevitable. Symbolic modeling lets a system construct a compact, reusable, highly generalizable mental model of a problem space using minimal data.

A recent NYT article by @CadeMetz [1] claims that neural network distillation was first developed in 2015 by a team of Google researchers including G. Hinton [2]. Not true! I published the technique in 1991 [3]. [2] failed to cite [3]. See the overviews [4][6].

The NYT article [1] states: "... distillation is an effort to copy the behaviour of the system, as opposed to copying the text verbatim." That's right! Distillation enables non-trivial forms of plagiarism [5].

Annotated References (easy to find on the web):

[1] Cade Metz, New York Times (6 July 2026). "American AI companies Say Chinese Copycats are Quickly Catching Up"

[2] O. Vinyals, J. A. Dean, G. E. Hinton (2015). Distilling the Knowledge in a Neural Network. arXiv:1503.02531. [2] did not cite the 1991 neural network distillation procedure [3].

[3] J. Schmidhuber (JS). Learning complex, extended sequences using the principle of history compression. Neural Computation, 4(2):234-242, 1992. Based on TR FKI-148-91, TUM, 1991.
See Section 4 on the "conscious" chunker and a "subconscious" automatiser which introduced a general principle for transferring the knowledge from one neural net (NN) to another. Suppose a teacher NN has learned to predict (conditional expectations of) data, given other data. Its knowledge can be compressed into a student NN, by training the student NN to imitate the behavior of the teacher NN (while also re-training the student NN on previously learned skills such that it does not forget them). In 1991, I called this "collapsing" or "compressing" one NN into another. Today, this is widely used, and also referred to as "distilling" or "cloning" the behavior of a teacher NN into that of a student NN. It even works when the NNs are recurrent and operate on different time scales.

[4] JS. Who invented knowledge distillation with artificial neural networks? Technical Note IDSIA-12-25, IDSIA, Nov 2025. https://t.co/w0WhIVGXQx

[5] JS. How 3 Turing awardees republished key methods and ideas whose creators they failed to credit. Technical Report IDSIA-23-23, Swiss AI Lab IDSIA, 2023 (updated 2025).

[6] @hardmaru & JS (2026). Munich 1991: the Roots of the Current AI Boom. With a preface by David Ha.

Marginal cost is not a detail. It is the whole thing. Imagine going to an airline, and they tell you, "yes, we can get you to 75% of these top destinations" -- great, but how much does it cost for each one? And how long to get there?

Unsurprisingly, all of the strong contenders on ARC-AGI-3 so far use this type of approach.

Codex for making a personalized daily digest:

Codex for creating the first video of a black hole: