Moonshot pauses new Kimi K3 subscriptions after GPU demand maxes out in 48 hours
Moonshot has temporarily stopped selling new subscriptions for its Kimi K3 model as demand nearly maxed out its GPU capacity within 48 hours. The company plans…
Latest
[Paper] xHC: Expanded Hyper-Connections – Scale Residual Streams Wider · Push Model Intelligence Further
Hyper-Connections (HC) expand the residual stream of Transformers into N parallel streams, providing a form of memory scaling beyond model width and depth. Manifold-Constrained HC (mHC)…
Every bit counts when you're reducing P(Doom)! Unexpected hope for AI slowdown from Prof. Gao:
Every bit counts when you're reducing P(Doom)!Unexpected hope for AI slowdown from Prof. Gao:Ivan Masli: @henrysgao Isn't that data centers are already in Western part of…
I'm Done with Mullvad
Article URL: https://moonwriting.blog/i-m-done-with-mullvad Comments URL: https://news.ycombinator.com/item?id=48975428 Points: 4 # Comments: 0
Models are still retarded at least they're starting to get better at self-correction we do need better models
Models are still retardedat least they're starting to get better at self-correctionwe do need better models
We're talking past our models; or, How a model defined its "evil" vector as dread
SummaryWe train a new token—a neologism (Hewitt et al.)—for a model, but unlike Hewitt et al., we train it on data the model generated while steered…
EU Exempts Apple Watch and AirPods from Battery Removal Requirements
Article URL: https://mjtsai.com/blog/2026/07/17/eu-exempts-apple-watch-and-airpods-from-battery-removal-requirement/ Comments URL: https://news.ycombinator.com/item?id=48975245 Points: 13 # Comments: 21
S1-Omni: A Unified Multimodal Reasoning Model for Scientific Understanding, Prediction, and Generation
We present S1-Omni, a unified multimodal reasoning model for scientific understanding, prediction, and generation. AI for Science (AI4S) has advanced significantly through domain-specific models, tool-augmented LLMs,…
Introducing Scylla’s Band, a new TTS model + inference framework with Android sample!
Hey all! https://github.com/lowkeytea/scyllasband -> inference code https://huggingface.co/spybyscript/scyllasband -> model, LiteRT, ONNX, and voices https://lowkeytea.github.io/scyllasband/ -> sample audio for the voices, emotions, and languages. The tldr: 10…
> China’s doctrine of total war applies to AI like anything else what does this mean the US has weaponized all of its supply chain advantages, is thi…
> China’s doctrine of total war applies to AI like anything elsewhat does this meanthe US has weaponized all of its supply chain advantages, is this…
> the Yangtze River Delta will continue to lead in technology and even expand its advantage over the Pearl River Delta Reminder: China is a big countr…
> the Yangtze River Delta will continue to lead in technology and even expand its advantage over the Pearl River DeltaReminder: China is a big country,…
China is basically taking away the WF6 market from Japan that's important for chemical vapor deposition I think the endgame is not just selling WF6 gl…
China is basically taking away the WF6 market from Japanthat's important for chemical vapor depositionI think the endgame is not just selling WF6 globallyTeo: The most…
b10069
opencl: Support broadcast for Adreno MUL_MAT and honor view_offs for Adreno Q8_0 MUL_MAT for llama-server multi-stream (#25910) opencl: handle broadcast for adreno gemm/gemv_noshuffle opencl: honor view_offs…
RT interstice: People in math-adjacent fields like economics shouldn't be so smug. Displaced pure mathematicians will invade and take over adjacent di…
RT intersticePeople in math-adjacent fields like economics shouldn't be so smug. Displaced pure mathematicians will invade and take over adjacent disciplines like the sea peoples in…
RT QC: for some background, 1) the jacobian conjecture is by far the most famous open problem resolved by an LLM so far 2) it's also infamous for attr…
RT QCfor some background, 1) the jacobian conjecture is by far the most famous open problem resolved by an LLM so far2) it's also infamous for…
I regularly forget that there exist pending DM requests I have ghosted like a 1000 people (and probably more bots) sorry everyone
I regularly forget that there exist pending DM requestsI have ghosted like a 1000 people (and probably more bots)sorry everyone
AI Morning Brief — 20 July 2026
Overnight in AI: Alibaba previews the 2.4T Qwen 3.8 Max while Moonshot AI suspends new subscriptions after Kimi K3 exhausts GPU capacity.
RAGU: A Multi-Step GraphRAG Engine with a Compact Domain-Adapted LLM
Graph retrieval-augmented generation (GraphRAG) enhances large language models with structured knowledge, yet existing systems construct knowledge graphs in a single extraction pass, producing noisy entities and…
Typical diplomatic supremacy of Kazakhstan: it's both in Pax Silica and in China's WAICO this is like Iran, India and Pakistan all being members of th…
Typical diplomatic supremacy of Kazakhstan:it's both in Pax Silica and in China's WAICOthis is like Iran, India and Pakistan all being members of the Shanghai Cooperation…
not much happened today
**US policy debates** are moving toward restricting Chinese open models like **Kimi**, with potential **procurement restrictions** and **Entity List designations**. Technical voices including **@APompliano**, **@ClementDelangue**, and…
I wanted to say that Mythos system card doesn't let us rule out 2x acceleration, but opened it and it has changed a bit. Interesting updates:
I wanted to say that Mythos system card doesn't let us rule out 2x acceleration, but opened it and it has changed a bit. Interesting updates:Ryan…
Reducing the model parameter size?
If the new releases of Chinese open weight models arrive at 2T+ sizes, is it possible for a research institution (with GPU clusters) to somewhat easily…
Concerning
ConcerningChristian Szegedy: I can't even fathom the next year
«Industry observers are already calling it a potential “DeepSeek moment” for the global energy storage industry.» lmao well, sodium will be a shoc…
«Industry observers are already calling it a potential “DeepSeek moment” for the global energy storage industry.»lmaowell, sodium will be a shock to someKyle Chan: CATL already…