G9v3-39A5B: Agentic heavy MOE with low hallucination
Hugging Face Artificial Analysis Should be a sweet spot for general work. Seems like coding is the only part that is inferior to Qwen. submitted by…
Full feed
Feed 28,389 posts
Every story across every category, newest first. Each card links to the original publisher; daily-brief posts open as editorial pages.
Ling-3.0-flash is another potential model to test before qwen3.8 27b
I tested Ling-3.0-flash with hard bugs and it fixed bugs that qwen3.6-27b could not. This models speed faster than deepseek v4 flash but almost the same…
LLMs reward expertise
Article URL: https://www.seangoedecke.com/llms-reward-expertise/ Comments URL: https://news.ycombinator.com/item?id=49161518 Points: 134 # Comments: 46
RT Modal: Congrats to Sakana AI on shipping Namazu! Happy to power Namazu's ~1T-param model for live web search + code execution on Modal.
RT ModalCongrats to Sakana AI on shipping Namazu! Happy to power Namazu's ~1T-param model for live web search + code execution on Modal.Sakana AI: 🐟 Sakana…
RT Modal: Congrats to Sakana AI on shipping Namazu! Happy to power Namazu's ~1T-param model for live web search + code execution on Modal.
RT ModalCongrats to Sakana AI on shipping Namazu! Happy to power Namazu's ~1T-param model for live web search + code execution on Modal.Sakana AI: 🐟 Sakana…
Why biological weapons are scary, and what we can do about it
What’s the biggest thing you think you can take in a fight? According to a YouGov poll, 6% of Americans reckoned they could beat a grizzly…
Constitutional Midtraining: Content Presence Drives Alignment Gains
Post-training alignment is often shallow, eroding under fine-tuning. Whether midtraining interventions, cleanly isolated from post-training, can produce durable alignment remains untested. We test this via constitutional…
Not only does Gemma 4 31B have 2.4x more active parameters and more computationally intense attention But for each token of context, it uses 13.3x mor…
Not only does Gemma 4 31B have 2.4x more active parameters and more computationally intense attentionBut for each token of context, it uses 13.3x more bits…
The one serious argument is that as a result of your GDP-maxing, you will now need a miraculous Wunderwaffe to win a war you're committed to have. The…
The one serious argument is that as a result of your GDP-maxing, you will now need a miraculous Wunderwaffe to win a war you're committed to…
One of my first bangers, aging like fine wine
One of my first bangers, aging like fine wineSusan Zhang: this is how these people somehow manage to rewrite history to Always Be On The Right…
Characteristics of a King, anon Little Rubio will foam at the mouth like the Chihuahua he is, defending hypocritical "values" and "Freedoms" Golden Do…
Characteristics of a King, anonLittle Rubio will foam at the mouth like the Chihuahua he is, defending hypocritical "values" and "Freedoms"Golden Don can just admit that…
💯
💯@jason: This is going down to $25m in under 18 months… then $10m and finally $3m
Wow
WowNick shirley: 🚨 Here is the truth from Ceuta, Spain:60,000+ migrants from Morocco’s border stormed the small town and are now hiding in the mountains and…
Wait, this story about Wenfeng leaving his car in Tibet is… actually true? I thought it's an urban legend in the spirit of Gigafeng memes. I know he'…
Wait, this story about Wenfeng leaving his car in Tibet is… actually true? I thought it's an urban legend in the spirit of Gigafeng memes. I…
RT Cursor: Cursor can now read, write, and act across your Google Workspace. New plugins give agents direct access to Gmail, Google Drive, Calendar, D…
RT CursorCursor can now read, write, and act across your Google Workspace.New plugins give agents direct access to Gmail, Google Drive, Calendar, Docs, and Sheets.
DeepSeek V4-Flash (284B MoE) at 33 tok/s single / 68 tok/s aggregate on 2× RTX 3090 + a used quad-Xeon DDR4 server — full config
Ran DeepSeek V4-Flash-0731 — the full official checkpoint, not a re-quant — on commodity used hardware. Sharing because I couldn't find anyone else publishing Ampere results…
[Paper] Towards Scalable Lifelong Knowledge Editing with Selective Knowledge Suppression
In this paper, the authors tackle continued pretraining without the risk of catastrophic forgetting, by identifying parameters which can safely be changed without risking identified concepts,…
How to Secure AI Agents, MCP Servers, and LLM Apps in Production
AI agents, MCP servers, and LLM apps break the core AppSec assumption that applications do what their code says. This guide walks through a practical see-fix-protect…
b10243
llama : allocate indexer cache only in "full" indexer layers (#26474) Co-authored-by: Stanisław Szymczyk sszymczy@gmail.com Website: https://llama.app macOS/iOS: macOS Apple Silicon (arm64) macOS Apple Silicon (arm64,…
There will be Whale multi-agent, inevitably. They won’t do a V4-Ultra as a model. They already said they are dissatisfied with its architecture and a…
There will be Whale multi-agent, inevitably.They won’t do a V4-Ultra as a model. They already said they are dissatisfied with its architecture and are working on…
‘I ran my own benchmarks on it’ seems to be pretty common comment around here. How about dedicating a thread for this and sharing?
Of course, the concern is that in the end, this thread will be fed into the models' training data, but I feel benchmarking isn't so open…
Deliberate Alignment Faking as a Defense Against Model Poisoning
I want to discuss and brainstorm a counterintuitive approach to AI alignment:Inducing alignment faking on purpose, to prevent the model from developing emergent misalignment.To prevent this…
AI's debt binge can't last, hidden borrowing reaches $1.65T
Article URL: https://fortune.com/2026/07/31/ai-debt-hypescalers-capex-capital-spending-hidden-borrowing-bond-issuance/ Comments URL: https://news.ycombinator.com/item?id=49160699 Points: 40 # Comments: 3
AWS is helping vibe-coding startup Superblocks, and the implications are big
AWS now allows vibe coding tool Superblocks to be embedded into the private clouds of AWS customers. It's another step towards decoupling apps from models.