fnands

21 ms·

271.
by fnands 3y ago
Your resource link points to a missing page. How much do the solution you build get transferred between clients? Or do you start from scratch every time you work for a client?
272.
by fnands 3y ago
Have you had a chance to look at the Direct Preference Optimization [1] pre-print? It seems like it might help get around RLHF which (as far as I can tell) is the really hard/expensive part of training the best of these models. They
273.
by fnands 3y ago
Lol, some residual answers from what I assume is distilled answers from ChatGPT: Q: "What is your favourite conspiracy theory?" A: "As an AI language model I don’t have personal preferences or biases so my responses will alwa
274.
by fnands 3y ago
The nominative determinism is strong with this one
275.
by fnands 3y ago
Nice. Thanks for the summary. So ~4x the size of the Pile, any idea how it stacks up in terms of quality to other big datasets?
276.
by fnands 4y ago
It seems you are right! Well, I hope System76 smashes it and inspires someone in Europe to give it a go.
277.
by fnands 4y ago
Tuxedo computers? They are assembled in Germany, which is a step below manufacturing I assume, but at least they design their computers themselves, as opposed to re-branding a white label product like System76 does/did.
278.
by fnands 4y ago
Legit was my first thought when I saw the headline. But apparently it's real
279.
by fnands 4y ago
Shhh, don't tell them! ;-)
280.
by fnands 4y ago
I bought a 64 GB and replaced the SSD with a 256 GB ones myself almost immediately. From what I've heard reading from an SD card is fast enough anyway, but if you regret the skew you buy, upgrading is not much of a hassle. That being s
281.
by fnands 4y ago
Australia? Otherwise, I ordered mine on a Thursday and it arrived the following Tuesday.
282.
by fnands 4y ago
I made a similar comment elsewhere, but yes, for now it seems Valve is purposefully not pushing it too hard, for now at least. I think a large part of that is ironing out the kinks that come with launching a new product, especially in a spa
283.
by fnands 4y ago
Flop is a strong word. I think they sold 1M in ~6 months, with a lot of supply side constraint. That being said, I they're nowhere near the tens of millions of Switches sold last year. But I don't think Valve was trying to. They&#
284.
by fnands 4y ago
Super exciting to see where this all goes. As consumers, having the big boys competing for our attention is great place to be.
285.
by fnands 4y ago
Yeah, I was basically filtering out jobs applications by whether or not I could see what effect ML was supposed to have on the functioning of the business, and importantly, whether or not they actually had access to the data to make it work
286.
by fnands 4y ago
Who would be the customer? Or would this be for research purposes?
287.
by fnands 4y ago
Yeah, I feel like I'm missing something here, but in the other thread the guy says you can get everything you need for 5nm from Alibaba, but afaik to get to such small transistors (5nm) you need EUV lithography, and is there anyone oth
288.
by fnands 4y ago
Yeah, probably something along the lines of normal console update times (4-5 years). But maybe do the Switch thing and release an OLED version?
289.
by fnands 4y ago
“Marginally Aerobatic” Gotta love the understatement there.
290.
by fnands 4y ago
Yes, it was a tiny source, and luckily emits alpha particles which are easy stop by even small amounts of material. > * Unless it gets dispersed somehow. Exactly, the real worry was that it would get crushed and would be breathed in, in
291.
by fnands 4y ago
It was tiny, I think something like this: https://www.flinnsci.com/alpha-source-polonium-210/ap8794/ They were letting undergrads handle it, so definitely wasn't a large source. Also, hey Chase! Funny to see
292.
by fnands 4y ago
Oh, reminds me a of a time some students lost a Polonium source in the physics lab while I was a Masters student. The HoD had them crawling around the building for days with a Geiger counter, but they never found it. Luckily it was a small
293.
by fnands 4y ago
I really like your pricing model (i.e. use forever + 1 year of support).
294.
by fnands 4y ago
Ugh, worked with Keras/tf last week for the first time in a few years. It's so hard going back to tf after working in PyTorch. It just feels so clunky, and progress was so slow.
295.
by fnands 4y ago
I guess this speaks to the way that people like binary thinking. so "ML is cool!" -> "AI will take over all the jobs" "Quantum computing is cool" -> "Quantum computers will replace all other computer
296.
by fnands 4y ago
Today? No. Tomorrow? Maybe. I think it will be an expensive luxury for another while, but commonplace in a decade or so.
297.
by fnands 4y ago
Yeah, I think there's a lot of hype around generative models, but in my mind only a few use cases that will really stick. Stock photos/illustrations is imho the most likely industry that these models will change.
298.
by fnands 4y ago
Very cool to see someone come back out of "industry" and to crack a problem. On a side note: the name looked familiar to me, and I realized I had spent a decent amount of time re-implementing his most cited (ML) paper a few years
299.
by fnands 4y ago
Wild.
300.
by fnands 4y ago
How?

More ›