theptip

47 ms·

841.
by theptip 2y ago
a) after you create the superintelligence is likely too late. You seem to think that inventing superintelligence means that we somehow understand what we created, but note that we have no idea how a simple LLM works, let alone an ASI that i
842.
by theptip 2y ago
Trouble is, in practice what you would need to do might be “turn off all of Google’s datacenters”. Or perhaps the thing manages to secure compute in multiple clouds (which is what I’d do if I woke up as an entity running on a single DC with
843.
by theptip 2y ago
This seems to be so broad a definition as to no longer mean anything useful. People in general are interested in capabilities or economic impact, and GPT-2 cleared no notable thresholds in those regards. I prefer the exact opposite approach
844.
by theptip 2y ago
I’m gently optimistic on the subject. When bias is encoded in the synaptic weights of a human brain it’s extremely hard to quantify. You can’t run an ablation experiment or try different combinations of similar inputs to determine if a judg
845.
by theptip 2y ago
> humans do not need 10,000 examples to tell the difference between cats and dogs The optimization process that trained the human brain is called evolution, and it took a lot more than 10,000 examples to produce a system that can differe
846.
I tried to use AI to read a book(mcfunley.com)
2 points by theptip 2y ago|0 comments
847.
by theptip 2y ago
I do wonder if a better antitrust strategy would be to pursue a portfolio of actions against second- or third-tier companies like this; the current strategy of going up against Meta, Google, Apple is sound in terms of raw size of antitrust
848.
by theptip 2y ago
The workflow of having AI suggestions pop up on the rest of the line is really nice. You can ignore when you know what you are trying to write, but the really low-friction interaction of “write a very rigorous comment and then let the LLM a
849.
by theptip 2y ago
But… why go to all that effort when you can just pop the capsule and land in a thing that is not full of highly explosive fuel? Elon had that idea of using rockets for passenger transport so I’m sure it’s in the plan, but for astronauts I d
850.
by theptip 2y ago
It’s an interesting paper, but my worry is that this is cherry-picking a surprising negative result. (The paper does not seem to discuss how many evals were considered to find this case.) The general expectation should be that some AI failu
851.
by theptip 2y ago
LLMs perform well on “human knowledge” like you are describing. They don’t lack knowledge of simple relational facts like these. Nonetheless it would be interesting to see the problem reformulated in purely mathematical terms, I suspect mo
852.
by theptip 2y ago
The canonical REST solution of query params to add nested fields gets you quite far: GET /myresource?extra=foo,bar sure you over fetch a bit if you have multiple accessors. But agreed, if you have highly nested data especially when acc
853.
by theptip 2y ago
The industry has used “authz” and “authn” to disambiguate for decades.
854.
by theptip 2y ago
There is an important point here; the curriculum needs to change, a lot. There is little point in writing essays any more. We’ll need to find other ways of testing knowledge acquisition. But also, we’ll need other menus of skills/knowl
855.
by theptip 2y ago
I think you are misunderstanding. This post is about new capabilities in GPT-4o. So the existing reasons for RAG may not hold for the new model. Unless you have some evals showing that the previous results justifying RAG also apply to GPT-4
856.
by theptip 2y ago
I suppose the question then is - if you finetune on your own data (eg internal wiki) does it then retain the near-perfect recall? Could be a simpler setup than RAG for slow-changing documentation, especially for read-heavy cases.
857.
by theptip 2y ago
Amazing! Going further, I was recently wondering why we don’t just build two-layer (or higher) freeways in areas that have heavy congestion. “Express” lanes with fewer exits perhaps, to minimize transitions if that’s a problem.
858.
by theptip 2y ago
I think this trend will cut both ways. I have seen engineers, including senior-level, who think their job is just to write excellent code. This breed is doomed. I have also seen juniors with amazing ideas and ability to iterate, who get tha
859.
by theptip 2y ago
This could be the thing that saves the US healthcare system; currently we pay 2x what a normal developed country would pay for the level of service provided, because there is a thick layer of paperwork, with insurance companies going back-a
860.
by theptip 3y ago
Transparency is good, but I think it’s also important to impose contractual liability and fines too. GDPR has a good model here; a data processor must list all of their sub-processors, AND have contracts with each that let them enforce tran
861.
by theptip 3y ago
This is a great example of the concept “ubiquitous language” from Domain Driven Design. Use language that your domain experts understand. If your users know about NACHA files, using other terms would mean they need to keep a mapping in thei
862.
by theptip 3y ago
The game tree is both broader and deeper in Slay, so much harder to do tree search. Also very irregular; in chess you have very few situational decisions (eg en passant, castling) whereas many decisions in Slay are one-off (events) that are
863.
by theptip 3y ago
Don’t agree with this take. I’m sure Meta would be fine with other vendors taking over the low-end, if that meant there was a vibrant platform they controlled (ie the OS). They will lead with flagship models to push the boundaries of what t
864.
by theptip 3y ago
The flex reply here is to say how much value you created with your small team. Everyone is familiar with Instagram being acquired for $1B at 13 employees, and that track record would be sufficient to carry you a long way even if you did not
865.
by theptip 3y ago
Agree with your point. I think “developers that do 90% of their thinking before they touch the keyboard are really good” is the actual correct inference. Plenty of good developers use the code as a notepad / scratch space to shape idea
866.
by theptip 3y ago
I’m not sure “rights” are the correct framing. The protesters surely expected to be fired. They traded their jobs for attention on an issue they cared about. The question of what is a legitimate protest, and when doing illegal things is tac
867.
by theptip 3y ago
Do you have references for the “intimidated and threatened” bit? Is there a claim it went beyond a peaceful protest?
868.
by theptip 3y ago
It’s a checksum; by decomposing every transaction into a double of (credit A, debit B) that must sum to zero, you catch random arithmetic errors. You can think of it as “conservation of value”, so you can’t just create money out of thin air
869.
Bell's Theorem(en.wikipedia.org)
3 points by theptip 3y ago|0 comments
870.
by theptip 3y ago
No. There are no “hidden local variables” (color in your example), and it really is “spooky action from a distance”. See Bell’s Theorem, this has been mathematically and experimentally proved: https://en.m.wikipedia.org/wiki

More ›