Wrote anti-cheat ml, do ML/AI at places you know off and probably use; systems ML person who still cares about cycles, graphics, compilers in the agentic era
Helsinki
Joined July 2015
- Tweets3.2K
- Following3.1K
- Followers335
- Likes14.7K
We could just have a thread on the best talks/ content you've seen there.
youtu.be/8X69_42Mj-g?si=cVeF… 1 HR absolutely fantastic talk ( some follow ups on the llvm dev meetings). I would love the potential of LLMs with a chemistry DSL!
λndres Mariscal retweeted
I built bbtree, the foundation of databases and filesystems, and the fastest concurrent btree.
Code: github.com/XiangpengHao/bbtr…
Blog post & demo: bbtree.xiangpeng.systems/
λndres Mariscal retweeted
Hi folks! I'm looking for my next tech role and would really appreciate help spreading the word. I have 10+ YoE and am open to Senior/Staff/Principal, Tech Lead, and Engineering Manager roles. An intro or RT would mean a lot. ❤️ #OpenToWork
[ Rust, C#, Python, JS, AI, Infosec ]
Maybe cause we are keeping it for our own stuff! I've been using a custom CAD dsl for some side projects
λndres Mariscal retweeted
🚨 [Token][ization] Paper Alert 🚨
Tokenization is a wildly understudied area of language modeling despite it having effects across all of NLP. Over the past ~8 months, 32 (!) tokenizer researchers put together the most comprehensive survey of the field. Check it out!
I think the interesting part is not that you cannot have a good career in AI / formerly Ml without participating in these highly concentrated networking orgies, but we can't deny that there is a huge concentration in rationalist communities and formerly in effective altruism.
Yeah but you're beholden to their classifier and this distinction won't matter . You're distilling the models capabilities into a classifier you could just as well use that dataset on a small LLM or other things
To be fair that means the median agent has been exposed to purely slop, human made slop. There is a real gap in quality between really good codebases, now whether that matters in making products faster and the true incentive from the market is a different story.
λndres Mariscal retweeted
In case you missed it, the full N64 programming video was made public last week! You can check it out here: teapaw.org/rfleury/status/2100359…
The next entry, dealing with interrupt handling, diagnostic rendering, and controller input, will go public today.
. With agentic coding, I think this approach will be a lot more common! gamedev used to do it quite a lot like it too, I remember we had everything from re-implementations of stdlib stuff too for our anticheat. TigerBeetle really resonates with good engineering imo
I would think so, but we are on a crunch cycle in the industry, when hiring picks back up it will be a disaster
. It is happening, but the budgets for agentic loops are wide open right now , so less skilled people are getting away with it . If we reach a point where token efficiency matters the ones that haven't done anything of consequence will not be able to compete. However so many seams are wide open and constantly being patched over, cause things being stable is an afterthought when code is cheap. Just look at how many more incidents and reliability metric dips we are seeing, true expertise is still valuable, but a heck of a lot harder to filter when everyone can do it now.
hard to control for post-covid, many things are demonstrably worse. Sadly it leaks to our personal life sometimes