I saw kids recklessly commenting tech stuffs they don’t really work on. I wouldn’t say such stupid things but probably the 18-yr-old version of me would, aka amplify a random idea thought for 0.5s. I stopped interpreting this as purely bad but see this as young energy
I always wonder why the younger, stupid version of me grow much faster and happier. I think it’s just lower expectation, less distraction, more energy, and actionability coming out of stupidity.
@syang2ng Randomness exist everywhere but the rate in this community is crazy. Acceptance can largely depend on the batch of your reviewers. But nvm guess these things are making acceptances less and less important/meaningful.
I like attending AI confs. But its reviewing process is just so random that you have unprofessional and unethical reviewers who just unreasonably want to reject your work by saying bs like 28% improvement is minor and your work is incremental to some completely irrelevant work 😅
@abacaj Great work! FYI we also include our reproduction of 19 LLMs (12 types) on https://t.co/wRRBuIbj3t. We also made a pypi package to evaluate LLM solutions HumanEval with more rigorous tests (https://t.co/drldG0giz6). :)
@teknium@algo_diver@csahil28@abacaj IMO data contamination issue is okay for HumanEval since most problems are designed from scratch and the ground truth are compressed on the github. However, it could be a big problem for other datasets where the majority of problems are directly from geeks4geeks …
@ShunyuYao12 Sorry to hear this. Code owners actually can do whatever they want to their repo as long as they comply with licenses of borrowed code and GitHub unless you add a license on your paper to avoid 3rd-party impl. But yeah it is not a big deal to ref the repo of the original author.