Same for the enterprise (non-tech I assume) world. When I was on the business side, we treasured any compute we could get that was not tied to corporate IT. Going through them would turn a 1 day fix into a 2 week endeavor. Product development would go from 1 month to 6 or more.
I haven't read the paper, but sometimes asymptotic improvements do not translate to real world improvements due to a large multiplicative factor in the complexity that gets factored out in the O() analysis. So the dataset required to see speed-up is impractically large.
I agree on all points. I have used Python for 15 years, Julia for 3 and reach for Julia most of the time for personal projects. I was really stoked when at work the only FOSS solver for our problem was in Julia so we wrote the rest in it for easy integration. The only thing I dread is having to look for a new package since the ecosystem can be quite fragmented.
I'm not proud of it, but my #1 "Oh shit" git operation is to just delete my local repo, reclone, and reapply the changes. Works really well for me 95% of the time. The rest I ask dev ops guy to help.
I get it's just an example, but are we really this far gone? Just watch a random movie and if you don't like it, pick another. This is such an extreme micro-optimization of a small experience.
There's also a decent amount of controversy around really young GMs. Basically that their parents game the system by choosing official tournaments with burnt out GMs with low ratings so they can get their norms easier. Mishra recently had a lot of backlash from top GMs with those types of accusations. If that's true, those players will likely never reach the top ranks, but who knows.
Have you considered RAG instead of using the entire document? It's more complex but would at least allow you to query the document with your API of choice.
I disagree. I see an LLM as less calculator and more as cheating. I think there's a lot of value in creating something entirely yourself without having an LLM spit out a mean solution for you to start from.
LLMs have their place and maybe even somewhere in schools but the more you automate the hard parts of tasks, the less people value the struggle of actually learning something.
I don't know exactly how flycast works but I've done similar things with other emulators and you take an action in the game (take damage, type something), then search memory for that value. In this case the ascii code for the letter typed. Keep doing this until you've narrowed down a single block of memory that holds everything you've done
Very Happy to see FIGS on the list. I almost got a chance to use it on a recent project but the csutomer in the end decided they valued accuracy over interpretability. I wonder if the recent KANs fit into this? They're more interpretable than other NN architectures. Also, Berkeley has had a few articles on interpretable methods (including FIGS) over the years but idk how they compare to what you've already implemented:
Yeah it's simply model error. All models from Linear Regression to LLMs have error. I guess because this type of error is in the form of deceptively reasonable human language, it gets a different moniker. It's also notably harder to quantify so it might warrant a different name.
I used to do this as a data scientist. Sometimes training jobs would take an awkward amount of time like 30 minutes or an hour and I'd get sidetracked doing something else. So I just had it beep when it was done like an oven. Prevented me from having to check back every now and then or missing an email. We didn't have good chat integration back then.
I agree but I need to find an alternative. I live in texas unfortunately and walking is miserable for 4 months out of the year. I should just move but can't currently.
My second semester of algebra had a section on Galois theory and I remember thinking it was abstract nonsense and I didn't get it. I'm actually interested in going through this to see if my perspective has changed.
There's a VST that creates embeddings of all the sound samples you have in a directory and then projects it down to 2d so you can visually explore and find sounds you're looking for. Similar sounds are near each other.
I always thought doing that with the entire spotify library would be amazing. Give me a low dimensional space to explore the library. Even cooler if the embeddings have similar geometric properties of language embeddings where I could do arithmetic with songs to find interesting combinations.
Not at all? Totally different programming paradigm and performance. Certain communities pull towards Julia a lot more than others. Mostly I've seen scientific fields that require HPC but don't want to do everything in FORTRAN and C. Paging Chris Rackauckas!
Good! Tools like this will only get easier to use and more convincing. AI is always going to outpace regulations but we need to try to prevent abuse like this and much more. People have killed themselves over leaked pictures like this.
"I am Alex Levy, an independent researcher and AGI enthusiast"
Background appears to be exclusively in UI/UX design which shows because the website is very nice. But forgive my skepticism on everything else the page talks about. Might get some VC $ for the enthusiasm but not seeing any firepower behind the claims.
I played 5000 songs through spotify just last year. I don't listen to the same songs over and over. It would cost an insane amount of money, time, and space for me to discover even close to the same amount of music without Spotify (or some other streaming service). I consider Spotify an insane value at most any reasonable price because I really love listening to and finding new music nearly instantly.