nabakin

11 ms·

91.
Microsoft wants to make Windows Copilot open automatically on cursor hover(neowin.net)
4 points by nabakin 3y ago|2 comments
92.
by nabakin 3y ago
I mean, I'm glad I know there's a bias in those blog posts now. Until this post, I didn't know a16z funded companies/projects let alone biased their posts toward them and I've read a few of their LLM blog posts.
93.
by nabakin 3y ago
Opportunity to write a paper
94.
by nabakin 3y ago
Also aren't claiming they are the best LLM out there when they clearly aren't like Inflection. Overall solid
95.
by nabakin 3y ago
I mean you can say the inverse about merchant fees and large purchases. I think Stripe's way of doing things is better because it's inline with the actual expenses they incurr. To their servers, there's no difference between
96.
by nabakin 3y ago
This has more of the article, but not all of it https://www.livemint.com/companies/visa-mastercard-agree-to-...
97.
by nabakin 3y ago
Cool paper. It's more independent than dense or normal MoE but I think it's still far away from the distributed training you're looking for because you still need a seed LM which is trained normally and when fine-tuning each
98.
by nabakin 3y ago
I don't think MoE allows for that either. You'd have to come up with a whole new architecture that allows parts to be trained independently and still somehow be merged together in the end.
99.
by nabakin 3y ago
The original video https://www.instagram.com/insta.beakk/reel/CtIzWvDgWzX/ and there are more of him
100.
by nabakin 3y ago
I think it's good to raise awareness of bad practices if you recognize them.
101.
by nabakin 3y ago
I have some serious issues with this company's PR. They say their Inflection-2.5 model is the world's best personal AI[0] which is a dumb claim to make considering it is done off of automated benchmarks which we know are flawed an
102.
by nabakin 3y ago
Agreed. It's ridiculous people have to resort to saying their question dumb to avoid being attacked by toxic commenters.
103.
by nabakin 3y ago
> the idea of a bot-only MMO would be really interesting Then you may be interested in Screeps: World
104.
by nabakin 3y ago
Any idea if it's used in Canadian English?
105.
by nabakin 3y ago
Also uses "optimising" which is British.
106.
by nabakin 3y ago
One of the reasons why I doubt Satoshi is Hal is apparent from these forum posts: if my goal was to stay anonymous, I would not immediately associate my real identity with my anonymous one. Hal is one of the first people to reply to this Bi
107.
by nabakin 3y ago
Not wanting the lifestyle that would certainly come with it like harassment, being recognized constantly in public, drawn into lawsuits, controversy, the market hanging on his every word, swatting, etc.
108.
by nabakin 3y ago
People are using throughput and latency differently in different locations/contexts. Here they are referring to token throughput per user and first token/chunk latency. They don't mention the token throughput of the entire 57
109.
by nabakin 3y ago
There's a difference between token throughput and latency. Token throughput is the token throughput of the whole GPU/system and latency is the token throughput for an individual user. Groq offers extremely low latency (aka extreme
110.
by nabakin 3y ago
I've been thinking the same but on the other hand, that would mean they are operating at a huge loss which doesn't scale
111.
by nabakin 3y ago
He compares burglars to insects, not Asians. Do you have other evidence?
112.
by nabakin 3y ago
This is promoting gun rights. Seems like he's comparing burglars to insects. What part is racist toward Central Asians?
113.
by nabakin 3y ago
And less than a week after the interview with Putin. The comments section on that video is ridiculous.
114.
by nabakin 3y ago
I don't think we'll see much of a change in how we're approaching an asymptote for the foreseeable future. In order to do something like that, you would need a significant innovation that disrupts the whole trillions of dolla
115.
by nabakin 3y ago
I don't think the size of the room matters much. The speed of the floor is determined by the speed of the person walking, otherwise they'll reach the edge of the floor and fall off eventually. Maybe you could make the speed of the
116.
by nabakin 3y ago
If we spontaneously start calling them ClosedAI, it's similar enough that people will still know who we're talking about. Maybe we should start calling them ClosedAI from now on
117.
by nabakin 3y ago
Over the past year or so various projects have made it possible to run LLMs on just about anything. Some GPUs are still better than others like Nvidia GPUs are still the best for token throughput (via TensorRT-LLM), but AMD GPUs are competi
118.
by nabakin 3y ago
Absolutely and that will be a major usability problem when LLMs get things wrong as they so often do
119.
by nabakin 3y ago
The whole system isn't an LLM. At the point they are asking for confirmation, they've already parsed the required information and handed it off to normal code. It's not going to change again. Ultimately, an LLM only has the c
120.
by nabakin 3y ago
There's a keynote which explains it pretty well. It's basically a cheap phone focused 100% around an LLM assistant

More ›