peidaqi
@peidaqi
Followers
142
Following
7
Media
40
Statuses
1K
Principal Machine Learning Engineer at Zynga/TakeTwo. Opinions are my own. Git: https://t.co/4FRqprd78R Linkedin: https://t.co/cKOkuHZyLB
Waterloo, Ontario
Joined May 2007
What if he pulls a new MacBook Air 12 from the envelope, and then pulls an iPhone Air out of the MacBook Air?
The biggest question of the day: Will Tim Cook pull an iPhone Air out of a manilla envelope? (Created with ChatGPT.)
0
0
0
Asking Gemini to calculate average stock price in May. Gemini calls my MCP server and pulls an entire year of data and all this data becomes context for the next call. Imagine how this would burn through tokens in YOLO mode. Seems it's still not there yet. 😂 @GeminiApp
0
0
1
Very interesting article about potential security risks with agentic systems and MCP! https://t.co/Jy8FbiYsbq
invariantlabs.ai
We showcase a critical vulnerability with the official GitHub MCP server, allowing attackers to access private repository data. The vulnerability is among the first discovered by Invariant's security...
0
0
1
So... it's Claude Code without cursor/vscode integration? 😂😂
Say hello to the @geminicli, a local CLI to help you build and maintain software with 1,000 free Gemini 2.5 Pro requests per day : )
0
0
1
Unfortunately in Canada in winter when the road is fully covered in ice and snow currently FSD absolutely doesnt work😭
what a ridiculous moment this is. this image might be in textbooks at some point. “tesla launches robotaxi” doesn’t even begin to do any justice in any meaningful way. kinda just the beginning of post driving civilization. second only to space because it basically rewrites
0
0
1
😂😂
The billionth repository was just created on GitHub and it is perfect in every way!
0
0
1
For "less common" language such as C#, it seems AI coding such as @ChatGPTapp doesn't work as well as, e.g. Python/JS. It has introduced bugs and failed at fixing them in my tests. Bias from training data? - I wonder how it works with more complicated languages like C++/Scala.
0
0
1
Great milestone in Open Source AI!
Two years in the making, we finally have 8 TB of openly licensed data with document-level metadata for authorship attribution, licensing details, links to original copies, and more. Hugely proud of the entire team.
0
0
2
Not uncommon for humans either to have inner monologue using a mix of different languages they speak. Almost like the brain automatically chooses the most efficient way to think similar to RL .
0
0
0
It’s interesting sometimes @ChatGPTapp generates a conversation title that is in a completely different language. It seems the title is generated not only based on what’s shown in the app, but also the hidden CoT process and language drift is still prevalent in CoT.
1
0
2
Huge news!! - this and WSL open sourced yesterday!
Full Windows Support for ROCm.
0
0
0
Wow this is great news!! Maybe finally a fix for various networking issues!
Big news: Windows Subsystem for Linux is now Open Source! 🎉 Download WSL, build from source, contribute fixes & features, and join its active development. Learn more: https://t.co/JzhrU4RAkx
0
0
0
"refactor AI features from the extension into VS Code core" - not sure what this exactly means but probably not a good idea. People love vscode because it's a lightweight editor. Adding AI will bloat the software and raise security/privacy concerns.
0
0
0
This is great!!
Want to learn the engineering details of building state-of-the-art Large Language Models (LLMs)? Not finding much info in @OpenAI’s non-technical reports? @percyliang and @tatsu_hashimoto are here to help with CS336: Language Modeling from Scratch, now rolling out to YouTube.
0
0
0
In 3D IMAX in a theatre near you 😂😂
0
0
0
Interesting stuff 🤔🤔
the era of giving an LLM a “prompt” and getting a “response” is over. today’s LLMs are NOT chatbots!!! to get the most out of o3, give it: 1. a goal 2. success criteria in return, what you get back is ***cognitive work*** cognitive work towards the stated goal, verified
0
0
1
Hmm… so an annual @ChatGPTapp Pro subscription can pay for a fully locally deployed DeepSeek R1…. 😂😂
Cost to run full size Deepseek R1 model down to $2,000. 🤖🤯 @gospaceport HT @wasnt__me_
0
0
0
This is fast!!
TRL v0.14.0 is out! This release includes DeepSeek's GRPO: the hottest RL algorithm in town 🔥 🏃 Run it with vLLM for blazing fast training 🚀 Scale to the moon with DeepSpeed ZeRO-3 Now every LLM can have its own "aha" moment!
0
0
0