Search: “llm”

Pure scaling of LLMs did actually hit a wall; neurosymbolic AI rescued it. AI didn’t hit a plateau; pure LLMs did — until people used harnesses and tools to complement their weaknesses. Which was alwa

Right now standard CoT is a one-way street. You generate forward, dump a pile of scratchpad tokens into the KV cache, and pray the model doesn’t hallucinate halfway through. On phones/laptops that cre

More on the pelican on the bicycle test from @simonw: https://simonwillison.net/2025/Jun/6/six-months-in-llms/ I uploaded the source here so it's playable in the browser, forkable etc. https://karpath

This wins the prize for sleazy misrepresentation. @mattShumer took my 2023 argument for hybridizing LLMs with symbolic tools – which is *exactly* what everyone does nowadays – and made it sound like I

Correct. And that is a major reason why they are better now than in 2022-2023 when we had pure LLMs. (Which is pretty much what I said would have to happen, in 2022) When you are a normal ChatGPT/Clau

For the past few years, humans have been doing “prompt engineering” to coax the best performance out of different LLMs. In this work, we explored what happens if we train an AI to do that job instead.

1. I never said LLMs were not useful. They are, particularly with all the bells and whistles that are being added to them. I use them. 2. A robot-rich future can't be built with AIs that don't underst

And eventually schools will (if they have any sense) probably withdraw or limit LLMs. 🦔Schools across the US are reversing years of technology-first classroom policies after studies show laptop and sc

Are LLMs really more important than fire or electricity? “Honestly, a ton of what we’ve developed in my lifetime amounts to scaling up the delivery of information and entertainment and the frictionles

DeepAgents Sandbox — a lightweight native Linux sandbox for deepagents It gives full isolation for safely running LLM/agent-generated code — no Docker, no VMs, no cloud dependency. Super fast and secu

Do you want to know why LLMs feel sharp on surface semantics but hollow on the fine-grained stuff? “From Tokens to Thoughts: How LLMs and Humans Trade Compression for Meaning” Come by the poster tomor

Let me say this clearly: LLMs cannot feel emotions. Emotions are evolutionary mechanisms. They push us to avoid danger or approach what is beneficial. We experience emotions because we are alive, and

LLMs are still not consistent judges of qualitative work, and small changes to how that work is presented affect outcomes. Better harnessing and methods (multiple judging runs with randomized orders,

🔴 BREAKING Speculative decoding just got real for local LLMs. llama.cpp merged speculative checkpointing (PR #19493). Expect 0-50% speedups on coding tasks with `--spec-type ngram-mod`. This means: •

From this perspective, Gemini is also worse than the original Bard. Sydney was the original sin of LLMs anthropormism, but also got the idea that AIs can sometimes be better with personalities right.

Yann LeCun, Executive Chairman of AMI Labs, explains why LLMs are mostly retrieving human knowledge rather than thinking for themselves: LeCun starts with why so many people misread what these systems

I am seeing claims around LLMs, specifically Astra having made serious progress on robotics. The tasks that are demonstrated are simple pick and place tasks with parallel jaw grippers. LLMs can do pla

This dismissal of AI skepticism is egregiously wrong, or at least deeply selective, @jeffclune. Was I wrong that pure LLMs would always be vulnerable to hallucinations? Was I wrong that pure LLMs woul

This work introduces diffusion-augmented LLMs, a new class of models. They first suggest that speculative decoding needs a separate draft model. Diffusion LLMs give up the quality of the model they re

Impressive tool to explore frontier AI capabilities. Martian's AI Frontier lets you compare 44 LLMs by measured cost, quality, and reliability, then see how routing and repeated sampling change the fr

I don’t like ELI5 as default for “make this easy to understand,” a great thing about LLMs is the ability to work as universal translators across topics Asking for generic dumbed-down explanations is w

チームのLLM開発が本格化してきたので仲間を募集します！ 人数を必要以上に増やさず、一人ひとりが大きな役割を担うチームを作っています。高い基礎力と知的好奇心、そして違った強みを持つメンバーが集まって、全体の目的に向かって高い主体性を持って動いています。そんなチームの一員になれます。 僕自身もこのチームでメンバーと仕事することが何より楽しいです。 LLM経験者はもちろん歓迎。未経験でも、しばらくLL

okay so everyone is talking about this but you guys realize it's not an LLM/foundation model or even a neural network, right???? it's probably just like some XGBoost or logistic regression algorithm o

Just saw that the LLMs-from-scratch repository passed 100,000 stars on GitHub! This is super cool and motivating. I am really happy to see that this open-source repo has helped so many people. Thanks

⚠️⚠️⚠️ Constitutional AI is not working. Aligning LLMs is not working. We need a different approach. If society doesn’t place its bets differently, we are screwed. Anthropic's Mythos created fake iden

Yet another paper argues that LLMs aren’t close to doing real discovery. MIT and Harvard argue LLMs are nowhere near doing real scientific discovery. They published a paper called “Evaluating Large La

There's been a talk about how LLMs are only advancing in verifiable areas like math or coding, but that isn't what the data suggests. As models have gotten better at that, they are also better at solv

During a Bloomberg interview, Yann LeCun (@ylecun ) explains why LLMs are limited in terms of real-world intelligence during a Bloomberg interview. 'Language is a very approximate, reduced, quantized,

LLMs training is opaque to end users and responses can be personalized for industry, locality, individual. The Trump regime openly controls who has access to the latest models. Government control of L

LLMs write fast single-GPU kernels. Ask for a multi-GPU one and they fall apart. ParallelKernelBench (PKB) measures how they fail by benchmarking against 87 problems pulled from real codebases includi

New course on serving LLMs efficiently -- how do you serve models to many concurrent users at low latency and reasonable cost? This short course is built with @RedHat and taught by @cedricclyburn. Eff

Current AI is bandaids all the way down; LLMs can’t play nicely with basic tools like databases and knowledge graphs, and you never know what you are going to get. It’s long time to face facts: LLMs h

In @latentspacepod podcast, I shared my view on video generation, world models, LLMs, agents, continual learning and where the next frontier is. 1. Video models get most of their intelligence from lan

June 2024: The latest general-purpose LLMs could not count the r's in strawberry. July 2025: The latest general-purpose LLMs get gold in the International Math Olympiad. May 2026: The latest general-p

Note that as of last summer Waymo said that LLMs/VLMs were experimental, and that their foundation model then could “process only a small amount of image frames, does not incorporate accurate 3D sensi

This post discusses how traditional caching mechanisms assume deterministic behavior (identical inputs producing identical outputs), an assumption that breaks down with large language models and espec

Sebastian Raschka shared an HTML-formatted table comparing active parameter ratios across different large language model architectures for improved readability and to avoid text truncation. The resour

1/ Excited to report we have a new paper out @Nature today! The bottom line: training data for LLMs does not just fall from the sky - it is created in the context of existing social political institut

'It is categorically *not* a victory for pure LLMs; it’s a victory for borrowing from classical AI and CS to move *beyond* pure LLMs.' ☝️ Compound AI Systems alllll the way downnn 🤩🤯🤩 Claude Code (sti

Really amazing dissection. LLMs hitting rarely. Tools list is amazing. 🤩🤯🤩 Claude Code (still not AGI but biggest advance since GPT-4) is the most neurosymbolic thing I have ever seen in my life. 53 s

Important Nature Neuroscience paper shows how humans differ from LLMs. Many people currently believe that humans are just next-word predictors, like LLMs. But this new paper by Zou, Poeppel and Ding s

One of the most important properties of LLMs that we take for granted is that newer, bigger models are just better at everything. The AI Labs are pouring effort into economically valuable fields like

The human brain🧠 is incredibly efficient because it only activates the specific neurons needed for a thought. Modern LLMs naturally try to do this too (> 95% of neurons in feedforward layers stay sile

Jeremy Howard proposes three key enhancements for large language models: incorporating latent space diffusion-like reasoning processes, adding a persistent recurrent state mechanism, and implementing

Hilarious (and maybe a little bit scary) that even in 2026 Marc Andreessen still hasn’t learned that LLMs don’t know how to reliably follow system prompts. don't feel bad, guys, the techno-optimist ma

Jerry Liu discusses a key limitation of state-of-the-art large language models at AI Dev Day '26, highlighting that while modern LLMs excel at building applications, they struggle with fundamental tas