Sabitlenmiş Tweet
JimZ
5.2K posts
JimZ
@moonares
Train deep neural nets since 2015. Weight lifting. Tesla talk and FSD analysis.
North Carolina Katılım Temmuz 2010
59 Takip Edilen3K Takipçiler
New in Claude Code: Code Review. A team of agents runs a deep review on every PR.
We built it for ourselves first. Code output per Anthropic engineer is up 200% this year and reviews were the bottleneck
Personally, I’ve been using it for a few weeks and have found it catches many real bugs that I would not have noticed otherwise
Claude@claudeai
Introducing Code Review, a new feature for Claude Code. When a PR opens, Claude dispatches a team of agents to hunt for bugs.
English
@ivanfioravanti wishful thinking. Compute is everything. It's fighting for capital game as there are already 4-5 perfomrant labs now while they mostly focus on large variation models
English
I bet the former Qwen leading team will create a new lab and even stronger models 🚀
English
🎶 I've been using voice mode to write much of my CLI code this last week
Can't wait to hear what you think.
Thariq@trq212
Voice mode is rolling out now in Claude Code. It’s live for ~5% of users today, and will be ramping through the coming weeks. You'll see a note on the welcome screen once you have access. /voice to toggle it on!
English
bf16 too big? here it is the fp8 version that u been expecting for! try to enjoy the benefits of our size and sparsity on ur device!
Qwen@Alibaba_Qwen
🚀 Qwen3.5-397B-A17B-FP8 weights are now open! It took some time to adapt the inference frameworks, but here we are: ✅ SGLang support is merged 🔄 vLLM PR submitted → github.com/vllm-project/v… Check the model card for example code. vLLM support landing in the next couple of days! Hugging Face: huggingface.co/Qwen/Qwen3.5-3… ModelScope: modelscope.cn/models/Qwen/Qw…
English
JimZ retweetledi
@novoreorx 一点见解: 先进化为超级个体，再统领 Agent 组建超级团队。前者决定了你的天花板，后者决定了你的规模化能力。二者相辅相成：自己是灵魂，Agent 是分身。✌️
中文
@davidasinclair What’s the least number of steps needed? That’s what I think most people would like to know.
English
7,000 steps/day is linked to ~50–70% lower mortality. You move!
jamanetwork.com/journals/jaman…
English
@MistralAI Had some high hopes for Ministral3-3B, did some benchmarks, and the reults fell short, especially week in tool-calling. Also surprisingly weak in vision benchmarks, some quite bad.
English
Introducing the Mistral 3 family of models: Frontier intelligence at all sizes. Apache 2.0. Details in 🧵
English
English
Our latest post explores on-policy distillation, a training approach that unites the error-correcting relevance of RL with the reward density of SFT. When training it for math reasoning and as an internal chat assistant, we find that on-policy distillation can outperform other approaches for a fraction of the cost.
thinkingmachines.ai/blog/on-policy…
English
Several of my team members + myself are impacted by this layoff today. Welcome to connect :)
English
Here is my very first Unprotected Left Turn Video.
youtu.be/vH3zQhzucjY
YouTube
Sawyer Merritt@SawyerMerritt
Tesla FSD Beta was released 5 years ago today. The video below is from the very first night of public testing. Watching FSD V14.1.3 rolling out wide today, so much has improved over the past 5 years. What a time to be alive.
English
Here is a highlight clip from this morning's dedicated FSD Supervised v14.1.2 Unprotected Left Turns video. I'm so impressed with what @Tesla_AI has done on this very complex traffic pattern. Safety critical systems are NOT easy to deploy to the masses. Don't underestimate the importance of this core @Tesla capability.
English
FSD v14.1.2, going to early access today, will debut a much awaited feature 🏎️💨
English
Elon: “FSD V14.2 rolls out in a few weeks and then 14.3 a few weeks later, depending on safety testing. There is so much change that we are carefully confirming each one.”
Elon Musk@elonmusk
Many more Tesla self-driving V14 improvements still to come, but wide rollout of V14.1 has begun. 14.2 rolls out in a few weeks and then 14.3 a few weeks later, depending on safety testing. There is so much change that we are carefully confirming each one.
English
This is interesting. Assuming 10GB of the 12GB is for the model, usually the model deployed is at INT4 , so this is a 20B model. Taking LLM as an example, 20B is the size that starts to having a taste of frontier model. With AI5, Tesla can definitely push the model size to another 2X~5X.
Whole Mars Catalog@wholemars
FSD 14.1 is about 12 GB
English