FREE IDEA! Take a human-authored codebase from <2022, have teacher LLM create a "sloppy" reimplementation of a single feature, then train student LLM to undo the slop. Should bias away from broad exceptions, blank returns, attribute/type checking, or low-q defensive coding.
@alexjc
-
Best Time to Code with AI Agents: Morning CET Insights
By
–
Absolutely, noticeable even via the API in other tools — for months now imho! Best time to code with agents is in the mornings CET.
-
Legal Challenges and Financial Barriers in AI Disputes
By
–
Yes, I said that already. But it's one thing to point out that it's a difficult legal situation that many do not want to (or cannot financially) engage in. It's another to berate them and tell them they are barking up the wrong tree when nothing else works in their favor either.
-
German IP Law Invalidates License Scope in New Mediums
By
–
This would be explicitly invalid in Germany in particular, as they encoded common practice into law! In other jurisdictions, as discussed, you'd have to argue the case based on caselaw, e.g. for new mediums unknown at the time. https://
cambridge.org/core/books/int
ellectual-property-licensing-and-transactions/license-grant-and-scope/
… -
Copyright Transfer Laws: Unequal Legal Treatment for AI Creators
By
–
Except you can't broadly transfer your Copyrights to unknown applications in many jurisdictions (EU+US). The creators would win in court if it wasn't a two-tier legal system: good rules for big publishers, bad ones for individuals.
-
Decentralized CLI tool for bot messaging and LAN control
By
–
@keet_io Any chance of a simple CLI for bots to send and read messages? Could be connected to existing account from the app, or a new terminal-only accounts… Need a decentralized interface to control bots on LAN and/or robust to cloud failures!
-
Integrating AI Innovations into Competition Rules
By
–
Yes, looked at those and other papers too — so much low hanging fruit there! The hard part is fitting those innovations into the context of a competition (what the rules should be); there was at least one attempt in doing that I think…
-
Token Undertrain and Variance in Language Model Training
By
–
Many papers! The tokens were significantly undertrained already last year, and the trend towards even fewer tokens likely just makes the variance worse! NanoGPT's small competition hides variance problems by gathering statistics over many seeds, IMHO just switch to medium size.
-
Are Claude Instances Shared Among Anthropic Staff Teams
By
–
I would be fascinated to hear if the Claude instances doing the coding and research are *shared* between the human staff… (Presumably the bug-fixing Slack bot is.)
-
Don’t Trust Verify: Research on AI Model Reliability
By
–
[1] Don't Trust: Verify (ICLR 2024) https://
proceedings.iclr.cc/paper_files/pa
per/2024/file/0a79ecda13603817de4cdfc68b417e89-Paper-Conference.pdf
…