jebarker

34 ms·

571.
by jebarker 2y ago
I think the data is way more important for the success of LLMs than the architecture although I do think there's something important in the GPT architecture in particular. See this talk for why: [1] Warning, watch out for waving hands:
572.
by jebarker 2y ago
I don't think anyone defines AI as "doing the thing that biological brains do" though, we define it in terms of capabilities of the system.
573.
by jebarker 2y ago
> What this tells us for AI is that we need something else besides LLMs Not to over-hype LLMs, but I don't see why this results says this. AI doesn't need to do things the same way as evolved intelligence has.
574.
by jebarker 2y ago
It's OK though, since throughput considerations don't matter as hardware is so fast now and your boss just wants you to get the thing shipped
575.
by jebarker 2y ago
My bugbear with dentists in the US (after living my first 30 years in the UK) is that they all continually hassle me to have my wisdom teeth removed. Said teeth have never caused me any problems and are all through the gums. I can only assu
576.
by jebarker 2y ago
If you do intermittent fasting you won't necessarily live longer, but it will feel much longer.
577.
by jebarker 2y ago
AlphaFold 2 wasn't Q-learning based. It was supervised SGD and the "evoformer" they introduced is very close to a transformer. So it's not exactly an LLM, but it's a pretty close equivalent for protein data.
578.
by jebarker 2y ago
Wait, was the parent comment talking about computers or ML? I interpreted it as the latter. The former I can get behind and I'd retract my statement!
579.
by jebarker 2y ago
> most significant advance since electrification I just don't see how this can be claimed at this point.
580.
by jebarker 2y ago
For me, it's The Grapes of Wrath. Simultaneously raised my bar for what I consider good writing and made me much more empathetic to the plight of those at the bottom of the socio-economic ladder through no fault of their own. As releva
581.
by jebarker 2y ago
Exactly, and well put. I have a hard time identifying those things that really make me happy without any external influence and an even harder time comfortably sitting with the idea that objective of my life should be personal happiness.
582.
by jebarker 2y ago
From experience, I can say that If you were conditioned from a young age to believe that achievement and status were important in life it’s an incredibly hard instinct to let go of. You’re totally right that it’s damaging for those around y
583.
by jebarker 2y ago
I didn't make a prediction, let alone a timeline - I just said it's plausible. I suppose I did say "we'll see" so I'm implicitly predicting within a few decades.
584.
by jebarker 2y ago
> LLMs have an additional problem related to training data. We are already throwing all the data we can get our hands on at them. I think it's plausible we'll see a breakthrough in data efficiency that helps here. Humans are an
585.
by jebarker 2y ago
I'm not really disagreeing (as GDP is a crude measure), but rather thinking out loud. I don't think my individual life satisfaction and optimism should be influenced by nation-state economics to the extent that that's what th
586.
by jebarker 2y ago
Sure, they could choose to work on that problem. But why do you think that's a more important/worthwhile problem than chip design or any other problem they might choose to work on? My point was that it's not trivial to make s
587.
by jebarker 2y ago
Because chip placement and the design of neural network architectures are entirely different problems, so this solution won't magically transfer from one to the other.
588.
by jebarker 2y ago
I think the real killer use case of AR glasses is when multiple people have them and can look at shared 3D content. The TV show "The First" (sadly cancelled) did a good job of showing use cases for this.
589.
by jebarker 2y ago
Thanks, I need to read more. I get how a semantic group could occur but it doesn't seem obvious to me that groups and group actions are necessarily the correct abstraction for representing semantic similarity.
590.
by jebarker 2y ago
Can you explain? I'm struggling to see how group theory is relevant here
591.
by jebarker 2y ago
> People gotta make games, again PC seems to be where more gameplay innovation occurs because the barrier to entry is much lower. I'd love to see Playstation make console development more accessible for a wider range of developers.
592.
by jebarker 2y ago
I agree with what you are saying regarding luck and opportunity playing a bigger role in people's success than is often appreciated. But I think it's ok to enjoy and appreciate the existence of a person in the world that was in th
593.
by jebarker 2y ago
I suppose if the only purpose of a company to exist is shareholder returns this is true.
594.
by jebarker 2y ago
Please tell me this is satire
595.
by jebarker 2y ago
Seeing the beauty of the correspondence between symmetry groups and Platonic Solids for the first time is a standout memory from my math undergrad. Groups are awesome.
596.
by jebarker 2y ago
> the claim that Buckley's version of Hallelujah is better than Cohen's (even the initial 1984 version) is controversial, at minimum :-) It is controversial, but shouldn't be. Not because Buckley's version _is_ better
597.
by jebarker 2y ago
Totally agree. Although if you happen to have the skillset to make lots of money you can use that to help more people. That gives a reason to tolerate bullshit jobs too.
598.
by jebarker 2y ago
Even if you don't agree with intellectual property, adding provenance to LLMs is a problem worth solving. A solution is likely part of solving "hallucinations" and probably reasoning about truth in the world.
599.
by jebarker 2y ago
I'm not ignoring your request, but I need to think about it a little as my day to day work isn't just pasting together APIs. If the only thing an engineer is allowed to do is paste together a fixed set of APIs then they may well b
600.
by jebarker 2y ago
The point is that if you understand how the high-level code you write gets translated to what runs on the actual CPU and memory then you write more performant high-level code by default the first time round. So you can get to market just as

More ›