dharma1

12 ms·

121.
by dharma1 3y ago
Better than most humans would have come up with! How does this not have some kind of a model of the world, with real world analogies like this?
122.
by dharma1 3y ago
So much untapped potential in AMD, and funny they keep failing at the software and geohot has to save them
123.
by dharma1 3y ago
it's reddit, but /r/LocalLLaMA/
124.
by dharma1 3y ago
fantastic. Will keep my 3090 busy for a while! "Furthermore, we note that our model is only trained with cross-entropy loss (supervised learning) without relying on reinforcement learning from human feedback (RLHF). This calls for furt
125.
by dharma1 3y ago
US only :/
126.
by dharma1 3y ago
Would sound pretty boring OCRd but I wonder if a trained ML model could do a decent job of score -> expressive midi (2.0) performance
127.
by dharma1 3y ago
Completed the game straight after it came out, fantastic game. Think they could make a killing making it multiplayer, but doesn’t look like it’s happening. Didn’t they announce they’re switching to UE5 for future games and not developing th
128.
by dharma1 3y ago
AirLink Wi-Fi stream with Quest is great, not much different to tethered cable. I think there is a case for local ML becoming more popular too, I could see nvidia making a Shield like box at some point with a mobile 4000 series GPU and good
129.
by dharma1 3y ago
If Berkshire sold out of most of their bank holdings, that should tell you something. Large banks are broadly fine for now (cough Credit Suisse cough), but there are a lot of small/midsize banks that are more vulnerable, enough to have
130.
by dharma1 3y ago
yes in an unusually tight spot due to the rapid, significant increase in interest rates (risk-free yield elsewhere) vs what they can offer to depositors - risking bank runs, which previously have been rare. If we zoom out, we could say it&#
131.
by dharma1 3y ago
Banks are in a tight spot. They have short term liabilities (depositors might want to pull their funds out whenever) and primarily long term assets (loans), except for the cash buffer they get overnight deposit rate for from central banks.
132.
by dharma1 4y ago
4090 is amazing, but very large card. 3090 is "good enough" for ML - same 24gb vram - and you can pick them up used for half the price of a new 4090. That's what I did. WSL on windows apparently decent, or native PyTorch, dua
133.
by dharma1 4y ago
why the serif headline font tho? Because it's classical?
134.
by dharma1 4y ago
100% this. Not seeing AMD make significant progress in ML against CUDA and other proprietary Nvidia libraries for years. The silicon is clearly capable but the software isn’t. Perhaps it’s time for leadership change in this division at AMD
135.
by dharma1 4y ago
Will be skipping this gen but PCIe 7.0 looks to have incredible bandwidth - https://www.tweaktown.com/news/87023/pcie-7-0-specs-announce... I’m wondering at 128-512GB/s over the PCIe bus could we finally get
136.
by dharma1 4y ago
Reckon they will (if not already) use 4bit or 8bit precision and may not need 175b params
137.
by dharma1 4y ago
Interesting to see the somewhat muted announcement from LeCun who has been (one might say, for good reasons) sceptical of “autoregressive LLMs” and the path to AGI requiring something quite different https://m.facebook.com/s
138.
by dharma1 4y ago
OK so don't need offloading at all for the quantised model - nice. In practice, how good is the 30B model vs 175B?
139.
by dharma1 4y ago
I’d love to run this on a single 24gb 3090 - how much dram / SSD space do I need for a decent LLM, when it’s quantised to 4bits?
140.
by dharma1 4y ago
“I’m going to forget you, Ben. :(“ That one hurt
141.
by dharma1 4y ago
Google
142.
by dharma1 4y ago
For GPU accelerated ML in Win11 (PyTorch w/ CUDA) - is WSL enough nowadays, or still dual boot to Linux?
143.
by dharma1 4y ago
Maybe a couple of freebies (with watermarks with link to your site for viral growth and license for personal use), then upsell with credits. You mentioned newspapers - I think you could charge a bit more for a full commercial license for th
144.
by dharma1 4y ago
How about a smaller, more meaningful but still well-funded ML company like Anthropic
145.
by dharma1 4y ago
Yes - I’m more hopeful of LLMs (or whatever replaces them over time) getting more truthful and accurate sooner than we get rid of SEO spam. I would also imagine SEO spam gets worse soon with LLMs.
146.
by dharma1 4y ago
Wonder what the compute cost for this ends up being. My guess is they will have to somehow rate limit it - I’m sure the rollout speed will also depend how much people end up using it. Maybe they will figure out some ad revenue - but I don’t
147.
by dharma1 4y ago
Backpropagation - Seppo Linnainmaa https://en.m.wikipedia.org/wiki/Seppo_Linnainmaa
148.
by dharma1 4y ago
He’s been working on machine learning long enough now to have some chance of success. It may go the way of his rocket ambitions (nothing comes out of it) but let the man try
149.
by dharma1 4y ago
Yeah sounded weird to me too, I don’t see why artificial intelligence would get deployed in human size units most of the time. The AWS bill won’t be for 5 Amys, and I don’t think people will “dial up” to order them
150.
by dharma1 4y ago
I agree with John that there are probably other people in his position (great engineers with a lifetime of experience and enough financial resources to focus on something like this for 5-10 years) who could have a go at this and have a reas

More ›