daveguy

31 ms·

691.
by daveguy 1y ago
Why would someone want a random block chain company to own 5% of the possibility of destroying their holdings? What if the company gets co-opted by some nation state actor? Bitcoin really seems like a terrible place to keep wealth.
692.
by daveguy 1y ago
So you hadn't read your reference before you read my post? If so, you would have known the only illegal chess move was a missed attack square between a castle. For the record I didn't see any of your response before I completed
693.
by daveguy 1y ago
Not sure how you don't see the difference between an LLM f'ing up how a single piece moves vs forgetting to hit the clock, accidentally touching two pieces or forgetting to call check. At least we agree and recognize that a mouse
694.
by daveguy 1y ago
Did you read those? These are the "illegal" moves listed: 5. Mouse slip 4. Forgot to call check 3. Accidentally touched 2 pieces, tried to fix it 2. Forgot to hit the clock button 1. Castle through attacked square So, the only one
695.
by daveguy 1y ago
> You can test this very easily with running old models w/ new cradles. They don't perform well at all. This is because neither the LLMs nor the cradles are intelligent. > They have to work together. Exactly. Because they ar
696.
by daveguy 1y ago
> It is showing the same test applied to a human. But you misrepresented the test with respect to humans. Humans who know how to play chess don't make illegal moves. > That said, humans who have written entire books on chess have
697.
by daveguy 1y ago
Humans who know how to play chess do not play illegal chess moves. Humans can learn chess in an afternoon and never make an illegal move again. The rules are pretty simple, and they are rules that every LLM has seen dozens of not hundreds o
698.
by daveguy 1y ago
Transfer learning is one of the biggest unsolved problems in AI. And we are nowhere near solving it or even understanding how to go about it from an algorithmic perspective. We will definitely see collapse of the current hype train before w
699.
by daveguy 1y ago
Well, they are a Google organization. Being backed by a $2T company gives you more benefits than just marketing.
700.
by daveguy 1y ago
I agree. I was pointing out that these anti-US-troop actions by Iran were related to prior conflicts / actions by the US. There was unlikely any consideration to downstream reactions which will endanger our troops. Completely short-sig
701.
by daveguy 1y ago
> explosively formed penetrators that killed 196 US troops... Well, it's a good thing Trump completely neutralized retaliatory action against US troops. /s
702.
by daveguy 1y ago
15 years is a lot better than immediately resuming enrichment. Which happened as soon as Trump scrapped the previous agreement. It still took them 7+ years to get to where they are now. So, it would have been 22 years, not 15. Trump is a sh
703.
by daveguy 1y ago
Really no more dangerous than a 480v, 300A level 3 charger in the US. Both have enough to kill. It's not like you'd be less dead if a level 3 charger malfunctioned. They both require redundant monitoring hardware and female plugs
704.
by daveguy 1y ago
Doesn't being able to only handle a single layer of abstraction as long as it is well defined and isolated essentially mean it can't handle abstraction at all? Aren't the layers the point of abstraction?
705.
by daveguy 1y ago
Also, regardless of the prompt they only get ~80% accuracy on coding benchmarks. So even with the absolute perfect prompt incantation, you can expect it to fail 1 out of 5 times.
706.
by daveguy 1y ago
It would be interesting to compare the iron nitride magnet sample with an just epoxy + iron filings. I would expect iron to have some area between the curves because iiuc iron is able to hold a magnetic charge.
707.
by daveguy 1y ago
> Mine is silent, I find it gross when you can hear someone's breath... Hate to break it to you, but you're in for an upsetting aging process. Also, your breath already isn't silent. Your brain attenuates the expected soun
708.
by daveguy 1y ago
> How would you reconcile this with the fact that SOTA models are only a few TB in size? Trained on exabytes of data, yet only a few TB in the end. This is false. You are off by ~4 orders of magnitude by claiming these models are trained
709.
by daveguy 1y ago
That was a sibling comment of the GP (sterilized mosquitos), the article itself is about a genetically engineered disease. The article has mosquitos "releasing toxic proteins in their semen". Seems like the sterilization is a much
710.
by daveguy 1y ago
> By that choice, I am left to assume this person feels they are the better half which says a lot about them. What a ridiculous assumption. Maybe they consider themselves and their partner to be equal halves of a whole. You know, the def
711.
by daveguy 1y ago
LLMs are trained with hundreds of terabytes of data to a few petabyte at most. You are off by 3 to 6 orders of magnitude in your estimate of training data. They aren't literally trained on "all the data of the internet". That
712.
by daveguy 1y ago
It could be. That's why we need more research. Well funded and not vilified.
713.
by daveguy 1y ago
Okay buddy. Enjoy your day.
714.
by daveguy 1y ago
Thread post. Not article post. Try to keep up.
715.
by daveguy 1y ago
And none of the items in the original thread post were about telemetry. They were about functional requirements.
716.
by daveguy 1y ago
Microsoft put ads in their OS menus. Kindof a stretch to "both sides" this one.
717.
by daveguy 1y ago
Excellent point. "Any" and "at least" were overstated. Should have been "all so far." But I'm definitely going with the side where we have all the evidence so far (even if it isn't proof) when decidi
718.
by daveguy 1y ago
> Many instances of NP complete problems are solved optimally all the time ... Okay, but the optimal solution of any NP-complete problem is still at least superpolynomial in complexity. If "optimally" also meant general-case
719.
by daveguy 1y ago
> That’s testing whether the model can emit a huge structured output without error, under a context window limit. Agreed. But to be fair, 1) a relatively simple algorithm can do it, and more importantly 2) a lot of people are trying to b
720.
by daveguy 1y ago
That's why doctors publish case studies all the time -- to inspire larger scale and statistically sound studies.

More ›