Unsloth studio, on a server that I built in 2024 that was, at the time, ludicrously oversized, but is now just ok-sized to run the q4 quant of qwen flash next. (128GB ram, 16GB vram).
I really want to build something less scuffed, but, yknow. So I'm making do atm.
There were a bunch of high performance computing efforts from the 1990s and on where, if you could work out that you'd need 5 years of "today's" compute, you should obviously hang on to the money for 4 years, then spend it on the correspondingly improved hardware and finish the job in a under a year. I've been considering building a box to run OLMO (one of the only models with training data "that you wouldn't have to defend in court" as far as I can tell?) and even with Sam Altman's supply chain attack, stalling for bigger hardware still seems like the right path...
That's a very interesting take. Unfortunately I can't wait with anything so I'll just use what's available today. But still you had me wondering for a bit.
Kind of funny title since I didn’t think UUIDv7 had any of the properties being disputed here. The reason to use UUIDv7 is because it reduces the space over which the UUID can cover at a given time, which improves write clustering. They don’t need to be strictly ordered for this, just “good enough”.
I hope to god no one is actually using the timestamp portion to replace an “updated at” timestamp but I suspect I’ll be disappointed.
Yeah this is what pick-up "artists" call a "neg" - a negative judgement made purely so the recipient feels, at some level, that their credibility is on the line. In this case, for clickbait. I always either don't read such an article or come here to read the comments instead.
I will not believe this until every frontier AI lab socializes a significant portion of their “profit” (subject to making a profit). I’m not talking about “Every kid with a Trump account gets a single share of SPCX”. I’m talking about “50+ percent of the shares are in a trust that distributes regular dividends to every US adult”. I’ll even be nice and allow those shares to be non-voting.
The problem is that such a line doesn’t exist, because they’re always looking for ways to increase profit for shareholders. On day X, they’ll be fine offering you an ad free sub, but on day X+Y there will be a meeting where they discuss what they can do to improve their revenue numbers and someone will suggest adding just a few ads to the ad-free tier.
This is why people go scorched earth with adblockers. There’s a lot of discussion about “good” ad networks being allowlisted, but they are all just one missed earnings report away from implementing dark patterns. None of them are trustworthy.
> Like first of all I would call a person who tries to pass the checks and balances of a democracy, even if they fail, the target of the ideals or “no kings”
Its incredibly frustrating to talk to people that take the position of "Well eventually they gave up after pressure from the courts and legislature, so no harm no foul." There's a difference between doing things that have a reasonable legal basis that the courts disagree with and what much of this administration was doing -- sometimes fragrantly ignoring court orders until D politicians got involved to pressure compliance.
Using the 2020 coup attempt as an example, yes, the courts shot down his attempts. The issue is that *he even tried* -- the attempt was transparently not good-faith concern about election integrity.
> And there's such little empathy out there for anyone going through it like me. AI is here now loser, keep up, get on board or get replaced.
I think that this really needs to be internalized by a lot of people. I really appreciated when my boss sat our team down and said "Yeah this is the new normal, and I know its not what you signed up for. Its fine to feel sad that your job is changing in such a dramatic way." No, it didn't help anyone actually adjust, but sometimes you just need someone to empathize and tell you its OK to feel bad about this when everyone around you is excited.
> If you read this far thanks. And if anyone has any advice for me, how to go from "thinking through problems by coding your way through them" to "doing the thinking upfront" then really, I'm all ears.
I'm trying to replace it with "thinking through with AI" to limited success. A lot of people recommend having the AI "plan" the thing and review the plan, which is OK. I've also had it do a prototype and poked around a bit, summarized the learnings, and then did it again in an iterative manner without rough edges I've found. Whether you are given the space to do that is up to your org -- its hard sell to "do it again" but I've found that the second attempt leads to better results since I can prompt my way around the problems. Sometimes this also leads to a "Not even wrong" moment where I found out that I framed the problem wrong and the AI dutifully carried out doing the wrong thing.
Work always comes with bullshit. Its the nature of the thing. I don't think anyone was really getting code golf-style experiences at work, but they likely had enough positives that they outweighed the negatives, which resulted in net "enjoying" work. For a lot of people, AI removed a lot of the positives and left a lot of negatives, and some are still coping with that in one way or another.
Sure, but how are you measuring this?
reply