bitL

18 ms·

211.
by bitL 4y ago
There are attempts to solve hard optimization problems in an instant with them.
212.
by bitL 4y ago
It's really terrible that Apple markets this as the next big thing but forgets to include detailed documentation so people have to experiment and figure out what works...
213.
by bitL 4y ago
News in 2030: Microsoft acquires Google for $50B.
214.
by bitL 4y ago
I think there is a typo, Shenzhen went from 88,000 (not 880,000) in 1985 to 11M in 2015.
215.
by bitL 4y ago
If that was true... You might end up with mass bacterial infections everywhere as only those cutting standards while providing highly addictive products survive etc. There was a reason why those regulations started initially. Overburdening
216.
by bitL 4y ago
Water pipe grid throughout the 3D piece?
217.
by bitL 4y ago
Oculus is not even being sold in Germany, the largest EU market.
218.
by bitL 4y ago
Glad to hear that! My 8-core AMD laptop from ASUS with 3080 is almost 2 years old now so things might have changed. My Threadripper still has USB issues and I don't like e.g. inability to type text or move mouse for 10-20 seconds just
219.
by bitL 4y ago
They could do what Microsoft used to do in the 90s-00s, make pirated Windows/Office available (by turning blind eye on private users) so that they capture and keep the mindshare.
220.
by bitL 4y ago
No, the original one. Didn't even know there was an upgraded one.
221.
by bitL 4y ago
Yes, Reverb G2.
222.
by bitL 4y ago
Fine-tuning is a solved problem using transformer adapters which are fast, small and match regular fine-tuning, so that's not an issue. As for smaller models, their usefulness will quickly wane when they will be seen as producing trivi
223.
by bitL 4y ago
That's highly unlikely. I think most people can't fathom how huge is the scale of GPT-4 and what computing power it requires. Even if 10x optimizations, 10x performance and 10x memory (highly unlikely in 5 years) that's not g
224.
by bitL 4y ago
I can understand that for outside-working hours tasks. But not for working hours where many companies really try to do the activity tracking-based payments like as if the consultant didn't have time allocated on working on their tasks
225.
by bitL 4y ago
Well, I have USB3 problems with all my AMD machines which makes e.g. attaching a VR headset reliably difficult. Random quickly repeated disconnects of devices due to a buggy chip from ASMedia/ASUS that AMD uses everywhere. I like TDP o
226.
by bitL 4y ago
Embedding to Speech interface ;-)
227.
by bitL 4y ago
Imagine you are an ML consultant with an activity tracker installed. The company is only going to pay you for the short fragments of time you are typing the source code and launching subtasks, but not for the duration of those tasks if your
228.
by bitL 4y ago
Yes, many of them used the training time for quiet deep thinking without worrying about getting paid.
229.
by bitL 4y ago
"How much of a productivity increase is that? Well jeepers, if you’re only doing 1/5th the work, then you are… punches buttons on calculator watch… five times as productive." So basically have GPT write the most interesting p
230.
by bitL 4y ago
Many companies don't pay ML engineers/researchers for the time spent training an ML model anymore.
231.
by bitL 4y ago
Yeah, but it fits on a single GPU. Now imagine it scaled across 1000 GPUs.
232.
by bitL 4y ago
I think you misunderstood how the generation of text works. For each new token it samples probabilities given previous tokens, not averages, then chooses some token from the top k as the next one with rules that penalize repetition of some
233.
by bitL 4y ago
LLMs without any online storage can be at best convincing liars. Combining them together with an actual retrieval/QA system (e.g. by first fetching exact answer via a QA model and then reformulating output via GPT) could start feeling
234.
by bitL 4y ago
It's still much worse than 1M context on 16GB VRAM with Reformer, but at the cost of inference speed. And you can use FlashAttention in your own models to get a more efficient/sparse attention now as well.
235.
by bitL 4y ago
3060 wasn't mid tier performance when high-end Arc was released anymore.
236.
by bitL 4y ago
Do you know the details? Are they having a question answering model hidden inside that evaluates context returned from search and then ask Bard to rephrase it and merge with its generative hallucinations?
237.
by bitL 4y ago
I am wondering if the blandness of Sundar wasn't one of the main reasons for his selection as a CEO so that there aren't any parodies of his public performances like with Ballmer/Microsoft when the company inevitably falters?
238.
by bitL 4y ago
Generative AI with ClipSeg/ControlNet is the main way to make Adobe products obsolete. No wonder they are pushing it into production ASAP but their investments to advanced tools they had edge over competition like e.g. intelligent back
239.
by bitL 4y ago
Imagine rent around 1000-1500 EUR, 42% taxes and inflation making basic food prices and energy go up 100-200% and you suddenly live paycheck to paycheck with no social life and no ability to own home. It's much better to live in e.g. D
240.
by bitL 4y ago
Haskell is cute until somebody in the team starts using the most advanced parts in some common task, destroying team's ability to understand what's going on and make changes; in a few years the author of such code will have the sa

More ›