有时候我觉得拿LLM做很多事情太浪费计算资源，但是仔细想想计算机系统里大部分资源都没拿来直接做“有意义”的事情，比如图形界面、比如WebView、比如3D游戏。
有时候我觉得拿LLM做很多事情太浪费计算资源，但是仔细想想计算机系统里大部分资源都没拿来直接做“有意义”的事情，比如图形界面、比如WebView、比如3D游戏。
\bLLM\b
All AI Pulse updates tagged "\bLLM\b".
14 signalsBerkeley professor Frank Coyle: "The LLM can't execute anything - it's just a probabilistic n...
Berkeley professor Frank Coyle: "The LLM can't execute anything - it's just a probabilistic next word predictor. All it can do is talk back to you very intelligently." He breaks down Anthropic's new Claude Certified Architect exam through real production scena
In the era of base LLM scaling (2022-2024), I believed the LLM line of research would reach a...
In the era of base LLM scaling (2022-2024), I believed the LLM line of research would reach a capability plateau (as later seen with base LLMs). In late 2024, after the o3 test-time compute demo, I changed my views: the new models were showing genuine fluid in
Don't waste 2 years learning to build LLMs like Claude & ChatGPT.
Don't waste 2 years learning to build LLMs like Claude & ChatGPT. Stanford just dropped a 2 hour course on how to build LLMs from scratch. • 00:00 - LLM tokenization • 25:44 - how LLMs decode user prompts • 35:40 - training pipeline of LLMs • 1:16:47 - LLM arc
The cheapest token is the one you never send.
The cheapest token is the one you never send. Most AI agents keep sending the same context back to the LLM. CodeMode keeps execution inside a Code Sandbox, sending only what's needed for the next decision. Fewer tokens. Lower cost. Faster agents. https:// data
Codex with RGB image inputs could do this.
Codex with RGB image inputs could do this. LLM agents are better than most people thought.
The ideal gas law and LLM training are the same equation.
The ideal gas law and LLM training are the same equation. Susskind's derivation starts with free energy: F = E - TS. Minimize it at fixed temperature and volume. Push through the calculus and pressure falls out as p = -(∂F/∂V) at constant T and N. For an ideal
A PHD STUDENT IN AMSTERDAM TURNED HER OBSIDIAN VAULT INTO A RESEARCH ASSISTANT THAT WORKS WHI...
A PHD STUDENT IN AMSTERDAM TURNED HER OBSIDIAN VAULT INTO A RESEARCH ASSISTANT THAT WORKS WHILE SHE SLEEPS. 3,200 notes. 47 papers in a week. 12 contradictions she did not know were sitting inside her own thinking. It started with Karpathy’s idea of the LLM wi
THE LOCAL LLM PC MARKET JUST GOT VERY REAL This video breaks down a category that barely exis...
THE LOCAL LLM PC MARKET JUST GOT VERY REAL This video breaks down a category that barely existed a few years ago: mini AI PCs under $2000 that are being sold as real local LLM machines. Not gaming towers, not cloud instances, but small boxes with Ryzen AI chip
A DEV TURNED A USED 2019 MAC PRO INTO A 35B LOCAL LLM SERVER BY PLUGGING IN ONE AMD EGPU AND ...
A DEV TURNED A USED 2019 MAC PRO INTO A 35B LOCAL LLM SERVER BY PLUGGING IN ONE AMD EGPU AND PUSHING THE RIG TO 52 TOKENS PER SECOND FOR UNDER $3,000 TOTAL he posts a video of the mac pro under the desk, one thunderbolt cable running to a Paladin eGPU enclosur
So I almost added - Home GPU rig And it might be nice for self-sufficiency but again I still ...
So I almost added - Home GPU rig And it might be nice for self-sufficiency but again I still don't think local LLM stuff even comes close to cloud either in quality or performance (speed) or cost Making your home self-sufficient is nice though, I have: - 2x Te
1/ We use LLM judges to scale up costly human evaluation.
1/ We use LLM judges to scale up costly human evaluation. But to trust an LLM judge, you need… human evaluation. Our new preprint tackles this circularity: "Metric Match: A Subset Selection Approach to Evaluating LLM Judge Reliability"
"At @databricks , we can do an enterprise-grade LLM migration now in 30 days ...
"At @databricks , we can do an enterprise-grade LLM migration now in 30 days ... or less. That used to take as long as 3+ years." Databricks co-founder Arsalan Tavakoli
Released last week, and already more than 4M downloads on HuggingFace alone This makes Gemma ...
Released last week, and already more than 4M downloads on HuggingFace alone This makes Gemma 4 12B the most popular encoderfree VLM by a large margin. In addition to being the first-ever general purpose LLM with encoderfree audio input!