akash

OSS & FP enthusiast, #rustlang tinkerer, INTJ

[Chubby♨️ @kimmonismus]

Kimi fixed all 15 critical bugs in 10 hours in a single prompt that GPT-5.6 and Fable 5 refused to fix because of guardrails.'
Respect for Chinese open source models is growing daily.

[Brivael Le Pogam @brivael]

Encore un scandale.

On accuse Elon Musk sans arrêt de se servir de Grok pour faire passer ses idées.

Pourtant, le @neutralityorg vient de sortir The Neutrality Project : une étude indépendante qui prouve l’inverse.

Sur 18 modèles IA testés, 54 positions sur 60 atterrissent à gauche du centre.

Moyenne globale : -0.41 (biais progressiste massif, surtout environnement et valeurs sociales).

Et Grok 4.5 ? -0.02. Le modèle le plus neutre de tout le benchmark. Le seul vraiment proche du centre.

Pendant ce temps, les autres boîtes injectent massivement leur idéologie de gauche dans des IA qui répondent à des milliards de questions par jour.

Manipulation des masses à grande échelle.

C’est absolument scandaleux. Ça devrait être un scandale mondial.

https://t.co/X1tSvJfDEY

[akash @_aka_sh]

@suni_code Engagement farming 😌

[akash @_aka_sh]

@Kimi_Moonshot Nice! Thats something

[Victor M @victormustar]

Before the week ends, let's acknowledge one of the most INSANE week ever for open AI, with 25+ notable open-weight drops across every modality:

🧠 LLMs

→ NVIDIA Nemotron 3 Ultra: 550B hybrid Mamba-MoE, only 55B active, 1M context, MMLU 89.1. NVFP4 variant claims ~5x throughput on Blackwell. First openly-weighted 550B hybrid Mamba-Transformer, closing the gap with frontier closed models.

→ Google Gemma 4 12B: fully open dense any-to-any (text/image/audio/video), 256k context, encoder-free, 140+ languages, AIME 2026 at 77.5. Shipped with a 23-checkpoint QAT wave (mobile ONNX + MLX). Most deployable model of the week.

→ StepFun Step-3.7-Flash: 198B sparse MoE VLM, ~11B active, SWE-Bench PRO 56.3. Apache 2.0.

→ Liquid AI LFM2.5-8B-A1B: edge MoE, just 1.5B active, 128k ctx, MATH500 88.8, MLX-ready. Best on-device option this week.

→ JetBrains Mellum2-12B-A2.5B-Thinking: their first open MoE, near-Qwen3-14B coding at 2.5B active. Apache 2.0.

🎨 Image gen (the surprise of the week)

→ Ideogram 4: their FIRST-EVER open weights. 9.3B flow-matching DiT trained from scratch. #2 overall behind GPT Image 2, top open-weight model on Design Arena + LMArena. Strongest open checkpoint for text-rich images, full stop. It has taste. Still can't believe this is open weights.

🔊 Audio & Speech (a breakout week for open TTS, 4 labs shipped)

→ Boson Higgs Audio v3 4B: 102 languages, 21 emotions, singing/whispering/shouting, sub-second TTFA.
→ RedNote dots.tts: the only fully continuous (no codec) open TTS pipeline, Apache 2.0.
→ Google Magenta RealTime 2: real-time music gen, <200ms latency, text+audio+MIDI. multimodalart ported it to PyTorch within hours with live ZeroGPU demos.
→ NVIDIA Nemotron-3.5 ASR: 600M streaming, 17x more concurrent streams vs Parakeet RNNT 1.1B.

👁️ Vision & VLMs

→ PaddleOCR-VL-1.6: SOTA document parsing at 1B params, Apache 2.0.
→ Baidu NAVA: 6.3B joint audio-video gen, best-in-class A/V sync, Apache 2.0.

🎬 Video, 3D & World Models

→ NVIDIA Cosmos3-Super: 64B omnimodal world model coupling action trajectories with video+audio gen, for Physical AI.
→ JD JoyAI-Echo: up to 5-min multi-shot text-to-video on LTX-2.3.
→ ByteDance Bernini-R + VAST TripoSplat (single-image-to-3D Gaussian splats, MIT).

[Elon Musk @elonmusk]

SpaceX has almost finished writing V1.0 of an in-house AI training stack in C that exact-maps to 220k GB300s with 800G NICs, making heavy use of pipeline parallelism and getting as close to bare metal as possible.

The potential speed improvement vs JAX for large training runs is over an order of magnitude.

[akash @_aka_sh]

@elonmusk Will it be open-sourced 🫣?

[Even Realities @EvenRealities]

You can now run your coding terminal from your glasses, wherever you are.

Terminal Mode is live.

#DevTools #BuildInPublic #coding

[Lxyv @wholyv]

wake up babe

another breach just happened

[akash @_aka_sh]

@hunterhammonds Add me as well

[Claude @claudeai]

Introducing Cowork: Claude Code for the rest of your work.

Cowork lets you complete non-technical tasks much like how developers use Claude Code.

[vik @vikhyatk]

2026 interview questions:

- you are in the middle of a refactor and the model says 8% context left before auto-compaction. what do you do?

- how do you decide which tasks to give to claude, codex gpt 5.2 xhigh, and chatgpt pro?

- tell me about a time you disagreed with an LLM

[akash @_aka_sh]

@anything LFG!

[Josh Barzon @JoshuaBarzon]

6000 Years of World History in 1 Picture

[akash @_aka_sh]

@HumansNoContext @grok how would you do it?

[Craig Brockie @CraigBrockie]

Scientists just cracked the multiple sclerosis code after decades of searching.

Two specific gut bacteria are triggering the disease, and they've proven it using identical twins and mice.

This changes everything we know about MS:

[Anton Shkuratov @shkrtv]

I don’t think you understand.

Instead of building a Saas, spending hundreds of hours finding PMF and then scaling to $50k/month, just create a gpt wrapper that generates business ideas and position yourself as an idea guy that screams how easy it is to run and scale a business without actually having done it yourself.

I don’t think you understand.

You can just sell a subscription to a business that generates business ideas. Proven, money making. They print

Go do it now

Now is the time

Comment “Steve jobs” and I’ll send you five $1,000,000 ideas. Most follow otherwise you won’t make it to $50k/month

[Kaizen D. Asiedu @thatsKAIZEN]

I’m in LA.

I support deporting people who are here illegally.

And I think some of the ways Trump and ICE are doing so are counterproductive.

Here’s why.

[akash @_aka_sh]

Litestream: Revamped https://t.co/v47kaseE9R via @flydotio

[Andrej Karpathy @karpathy]

We're missing (at least one) major paradigm for LLM learning. Not sure what to call it, possibly it has a name - system prompt learning?

Pretraining is for knowledge.
Finetuning (SL/RL) is for habitual behavior.

Both of these involve a change in parameters but a lot of human learning feels more like a change in system prompt. You encounter a problem, figure something out, then "remember" something in fairly explicit terms for the next time. E.g. "It seems when I encounter this and that kind of a problem, I should try this and that kind of an approach/solution". It feels more like taking notes for yourself, i.e. something like the "Memory" feature but not to store per-user random facts, but general/global problem solving knowledge and strategies. LLMs are quite literally like the guy in Memento, except we haven't given them their scratchpad yet. Note that this paradigm is also significantly more powerful and data efficient because a knowledge-guided "review" stage is a significantly higher dimensional feedback channel than a reward scaler.

I was prompted to jot down this shower of thoughts after reading through Claude's system prompt, which currently seems to be around 17,000 words, specifying not just basic behavior style/preferences (e.g. refuse various requests related to song lyrics) but also a large amount of general problem solving strategies, e.g.:

"If Claude is asked to count words, letters, and characters, it thinks step by step before answering the person. It explicitly counts the words, letters, or characters by assigning a number to each. It only answers the person once it has performed this explicit counting step."

This is to help Claude solve 'r' in strawberry etc. Imo this is not the kind of problem solving knowledge that should be baked into weights via Reinforcement Learning, or least not immediately/exclusively. And it certainly shouldn't come from human engineers writing system prompts by hand. It should come from System Prompt learning, which resembles RL in the setup, with the exception of the learning algorithm (edits vs gradient descent). A large section of the LLM system prompt could be written via system prompt learning, it would look a bit like the LLM writing a book for itself on how to solve problems. If this works it would be a new/powerful learning paradigm. With a lot of details left to figure out (how do the edits work? can/should you learn the edit system? how do you gradually move knowledge from the explicit system text to habitual weights, as humans seem to do? etc.).