I think in general these smaller applications of AI are seriously overlooked even though they will actually add lots of economic value for companies and lots of quality-of-life features for users.
Huh, when I tried it WASD didn't work like I expected, hence I went back and checked and saw that it was ZQSD. Looks like they might have fixed it now since I posted. Either that or the lagginess caught me up; it takes a second before it starts moving forward, it might be I got caught up by that, went back to look at the instructions and noticed the ZQSD, then went back and used the arrow keys since that was easier.
edit: just checked the code, and they cover both layouts by using the physical key code rather than the character. And that was committed before I tried it. So it must have been the fact that it was a bit slow to respond that got me.
oh please... lesswrong was full of idiots already in 2010 when people outside of that community were laughing about "self-taught expert" Yudkowsky's bullshit physics takes.
They are not "further along" the AI discussion, they are a bunch of wackos LARPing as scientists living out their personal sci-fi scenarios.
what are you talking about bro, I think you're really getting carried away here. Nobody cares about the direction of writing on the spine of a book. And your thing with sentence vs title case doesn't make sense because German doesn't even have that distinction.
I currently run pi agent in Lima on a Mac with only the code project folder mounted and an extension that prevents pi agent from reading the contents of .env files directly.
Yeah, there probably are some freak situations where this isn't safe enough, but I don't really see any realistic ways this is going to end up badly. Am I overlooking some obvious security holes?
I designed it to provide a single interface to agent sandboxing, no matter how far up the security tower you want to go.
It eliminates the manual process steps you end up doing with an ad-hoc system (which gets old the 10th time you do it).
Common weak points:
- The agent can access your homedir.
- The agent can access .gitignored files, which can contain secrets (and are gitignored for this reason).
- The agent has r/w access to your workdir.
- The agent could follow your remote mounted dirs.
- The agent can act in your name with whatever credentials it finds (and it will use them when it tries to be helpful, especially with the gh tool).
- Do you even know what's in the diagnose_problem.sh file it just created and asked permission to run?
- Even the .git dir can be weaponized, such as with evil filters.
- The agent can edit its own process, bypassing the harness controls and giving it the same access as you have (amplified by each credential sitting on that machine).
Meanwhile, you're reflex-hitting ENTER without looking because 99% of the permission prompts are mundane.
I don't know, I don't think this "effort for effort's sake" is a very convincing argument. In particular, I think it's very much affected by recency bias in a way?
What we perceive as "effort worth taking" instead of "dull occupational therapy" is very prone to change with technology.
If you would argue that modern photographers need to take the time to physically develop their photos and use chemicals to get their effects rather than applying photoshop filters, you'd not be taken very seriously - in the 80s and 90s it would have been a very different discussion where people saw photoshop as "taking the helicopter to the summit of Mt Everest".
Same even with paper writing. I still had old school teachers in the 90s and early 2000s who insisted that writing anything on a computer was a "shortcut" that would encourage worse writing because you could undo stuff etc. They did all their handouts and worksheets on their old typewriters.
There is a discussion to be had on AI in maths, but I don't think it's this one. I think mathematicians should be talking about what the future of their field is supposed to look like in a time where AI will be able to find the proofs. Maybe maths will turn into a more "experimental" science, where you already know the proof of a theorem, but you want to find a particularly elegant way that helps humans understand it or find other ways to apply the knowledge. Or rewrite old theories from different angles based on all the new proofs generated by AI. I don't know, but I think there's a lot of mathematics to do out there for humans even in a time with AI.
I'm not arguing that people should do things as hard of a way as possible, I think that's kind of a straw man. I am not saying we should tie stones around us, write things in Latin, and not use computers or something like that.
Math is not really about just piling up results in my opinion. Most people's day to day lives were not directly changed by the recent Erdős problem that was solved by AI. I think most people believe the result would have been more impressive if a human had found the counterexample. People do not list those problems because they need to be urgently solved but because they are hard and interesting and it impresses us when a human solves it.
But not only that, the person gains insight along the journey. I am not sure what insight AI can give us really. It basically maximizes text output over a massive dataset that includes mathematical research. So again, the journey itself is valuable not just because it's hard but because it gives us information.
There's also the problem that credibility for so many things is based on attribution of doing something difficult. A PhD has been a way we measure credibility in a field. It's a fence you hop to get in. If AI can write up a thesis in a tiny fraction of time and cost, it could break attribution if people print theirs with AI but also it breaks the credibility link. Is that important? For math, using PhDs as a measure of credibility in some ways is a bad way to do things, but it's effective and probably the best we have. (For other fields it could be catastrophic). So it destroys a signal about the author which may not be catastrophic here. Where this goes wrong is if getting a PhD is now obsolete. People who have experience and who can review AI output proofs exist right now but how will we ensure anyone is trained in proofs in the future? If we're going to outsource math to AI I think the reason people are right now getting PhDs could be completely obsolete unless we agree it's worth it for people to work through it on their own and gain that experience for themselves.
I'll do my best to restate the problem. If AI does all the work, then we won't have people who can understand the math that AI's are creating.
This is the real problem as I see it and it extends to most technical work.
I didn't see that as the main argument of the article.
I was objecting to the old guard defending their turf.
Academia needs to contend with this problem, and start addressing it right now.
People are not going to be able to beat AIs at math. So whats the solution? PhDs will have to be able to show competence in new ways. Publishing and review of mathematical work will have to change. The genie is out of the bottle and there is no putting it back.
we did some internal tests.
The quality isn't bad, it works quite well. But it's essentially on the same level of an ARIMA model trained on the data just much bigger and slower.
So in my opinion it currently falls into a kind of void. If your use case is worth predicting and you put a data scientist on it, you're better off just training cheaper ARIMA models.
That is disappointing. One would say that with all the budget and compute, Google would be able to create something that beats methods from 70s. Maybe we are hitting some hard limits.
Maybe it would be better to train an LLM with various tuning methodologies and make a dedicated ARIMA agent. You throw in data, some metadata and requested window of forecast. Out comes parameters for "optimal" conventional model.
i met an associate working for a particular VC and they were really into time series foundational models. I argued the most of the "Why real forecasting problems break the whole frame" as to why they were wasting their time at that time.
she was totally convinced i was wrong because she was discussing investing with some top and well respected researchers that were really pushing this and wanted to make a startup around it.
i was and am still confused as at all the wishful thinking. then again, sometimes the best time to sell an idea is right before you think it is possible.
this completely misses how crazy word2vec is. The model doesn't get told anything about word meanings and relationships and yet the training results in incredibly meaningful representations that capture many properties of these words.
And in reality you can use it in much broader applications than just words. I once threw it onto session data of an online shop with just the visited item_ids one after another for each individual session. (the session is the sentence, the item_id the word)
You end up with really powerful embeddings for the items based on how users actually shop. And you can do more by adding other features into the mix. By adding "season_summer/autumn/winter/spring" into the session sentences based on when that session took place you can then project the item_id embeddings onto those season embeddings and get a measure for which items are the most "summer-y" etc.
mh, maybe it's cheating because it's still a STEM degree but I have a PhD in physics without any real computer science courses (obviously had computational physics courses etc. though) and I managed to 100% solve quite a few years without too much trouble. (though far away from the global leaderboard and with the last few days always taking several hours to solve)
