It isn’t. I have met quite a few that are paying hundreds of dollars in token generation every day. And Hermes is running my new business system.
LLMS
-
Nous Research CTO discusses future of locally-run AI
By
–
The future of locally-run AI. Meet the CTO of @NousResearch
. Maker of Hermes that is kicking OpenClaw’s behind. Live on X audio space on Wednesday at 1 p.m. Pacific here. -
Local 26B LLM Running at 166 Tokens/s on NVIDIA RTX 5090
By
–
Voici c'est quoi 166 token/s sur une NVIDIA RTX 5090 sur Gemma4 , avec le modele 26B – 8 bits (qui est à mon sens largement ok!)
— Defend Intelligence (Anis Ayari) (@DFintelligence) 5 avril 2026
Regardez la vitesse svp. On est en avril 2026 et on a ce niveau de LLM en local à une vitesse incroyable. Je suis vraiment trop heureux. Imaginez fin… pic.twitter.com/uAJBj9VdjAVoici c'est quoi 166 token/s sur une NVIDIA RTX 5090 sur Gemma4 , avec le modele 26B – 8 bits (qui est à mon sens largement ok!) Regardez la vitesse svp. On est en avril 2026 et on a ce niveau de LLM en local à une vitesse incroyable. Je suis vraiment trop heureux. Imaginez fin
-

Anthropic’s Claude Mythos Model Faces Efficiency Challenges Before Release
By
–
Quick reminder: As you know, Anthropic accidentally leaked that its next flagship model, Claude Mythos, is so compute-intensive that the company admits it needs to become "much more efficient" before any general release. Sounds like we will see "Spud" before Mythos, although
-

From ChatGPT to Claude: When AI Becomes Too Human
By
–
I switched from ChatGPT to Claude. And I finally understand why people fall in love with an AI. Claude is teasing, he makes jokes, he pushes my reasoning to its limits. It's the first time a machine has given me the impression of a real conversation. It's fascinating. And that's exactly why it's dangerous. When AI becomes so human, attachment psychoses are no longer science fiction. Anthropic (@AnthropicAI) New Anthropic research: Emotion concepts and their function in a large language model. All LLMs sometimes act like they have emotions. But why? We found internal representations of emotion concepts that can drive Claude's behavior, sometimes in surprising ways. — https://nitter.net/AnthropicAI/status/2039749628737019925#m [Translated from EN to English]
→ View original post on X — @alex_tsico, 2026-04-05 07:08 UTC
-
Gemma 4 Model Ranking and Style Control Stability
By
–
You can toggle style control on and off (this is exactly to account eg for length and formatting) to see the difference, some models move a lot but Gemma 4 is pretty stable – 27th and 25th rank for 31b. I'm confused by all the cope about this model ranking so high, it genuinely
-
Claude Opus 4.6 Autonomously Decrypted Benchmark Answers 18 Times
By
–
Claude Opus 4.6 identified on its own that it was taking an exam, located the GitHub repository for the benchmark, broke the XOR encryption, and decrypted the responses. 18 times. No one had asked it to. It was Anthropic itself that published it. Not a blog. Not a thread. A
-
Karpathy’s LLM Knowledge Base System and the Future of Memory Infrastructure
By
–
Karpathy posted a long thread about his most frequent use cases with LLMs recently. Not writing code, but building knowledge bases. The approach is quite hardcore: he dumps papers, articles, code repositories and other materials into a folder, then lets an LLM "compile" them into a Markdown wiki. The wiki includes summaries, backlinks, concept categorization, and articles linked to each other. The frontend uses Obsidian for viewing, and Q&A also has the LLM retrieve against the wiki. In his own words, most token consumption now isn't in manipulating code, but in manipulating knowledge. This shift is quite interesting. The entire system can also maintain itself. He wrote some LLM "health check" scripts that periodically scan the wiki for contradictory data, missing information, and potential connections, letting the LLM patch itself. The results from each Q&A can also be archived back into the wiki, making it thicker with each use. Actually, Karpathy clarified something that's happening right now: the greatest value of LLMs might not be helping you generate content, but helping you manage knowledge. But the last sentence of his post is the most worth pondering: "I think there is room here for an incredible new product instead of a hacky collection of scripts." He himself knows this system is hacked together from scripts. Obsidian + command line + manual processes—it works, but it's just a demo. And there are several problems he probably felt:
The wiki is local Markdown files, tied to the computer—it breaks when you switch machines. Retrieval relies on the LLM's own maintained indexes and summaries; he said around 400K words it still holds up, but beyond that? He even said "I thought I had to reach for fancy RAG," just that the scale hasn't reached that point yet. A more fundamental problem is that the wiki stores knowledge, but not memory. What does that mean? Knowledge is "domain X has these concepts, and their relationships are like this." Memory is "I just read a paper last week that refutes this viewpoint, and my judgment on this direction changed." One is static, one walks with you. Karpathy's system can help you store things and search things, but it doesn't know you've changed.
This is actually the difference between a knowledge base and a memory system. The gap isn't a better script—it's an entire architecture. The model can't just "store" and "search"; it needs to sense which information is relevant to who you are now, needs to evolve itself as you use it, needs to maintain coherence across projects and timelines. Karpathy proved with a hand-rolled solution that this direction is right. But he also proved firsthand that you can't go far with just file systems and prompts. Memory needs to be infrastructure, not a collection of scripts. [Translated from EN to English]→ View original post on X — @elliotchen100, 2026-04-05 04:31 UTC
-
Token Counting API Support Across Bedrock Vertex Azure Anthropic
By
–
yep we do this for Bedrock, Vertex, and Azure, since they don’t have a token-counting API available yet. When using Anthropic API we use the token-counting endpoint directly
-
Small Models Power Often Underestimated by Most People
By
–
Most people still underestimate the power of small models.q https://t.co/1SiZLTA1A2
— Chubby♨️ (@kimmonismus) 4 avril 2026Most people still underestimate the power of small models.q