theptip

46 ms·

691.
by theptip 2y ago
Aka “Ready, fire, aim” from the Pragmatic Programmer. This approach is fine but make sure your iterative prototyping is refining some sort of e2e / acceptance test, with good comments / README describing the intent. It’s really ea
692.
by theptip 2y ago
Yeah, I think for the highest-value activities (eg legal advice) you expect to run the full frontier model. But maybe you want to run a smaller one locally on your iPhone for privacy and accept the capability loss.
693.
Explore Theory-of-Mind: Program-Guided Data Generation(ai.meta.com)
2 points by theptip 2y ago|0 comments
694.
by theptip 2y ago
For a “legal LLM” you need three things: general IQ / common sense at a substantially higher level than current, understanding of the specific rules, and hallucination-free recall of the relevant legal facts/cases. I think it’s re
695.
by theptip 2y ago
> what should we make of the remarkable ratios found… We also can’t implement those creatures’ control systems in silicon, so they too are doing things we can learn from?
696.
by theptip 2y ago
As a modeling exercise: this isn’t necessarily inconsistent. Right now if Google can earn more from using their TPUs than they would selling them (ie profitably utilize them) they would be crazy to sell them. Companies are valued not on the
697.
by theptip 2y ago
Leaving aside the “AGI on paper” point a sibling correctly made, your point shares the same basic structure as noting that any VC investment is a terrible deal if you only 2x your valuation. You might get $0 if there is a multiple on the li
698.
by theptip 2y ago
$10B of compute credits on a capped profit deal that they can break as soon as they get AGI (i.e. the $10T invention) seems pretty favorable to OpenAI.
699.
by theptip 2y ago
As the article states, YC’s key asset is their deal flow. You simply don’t get that if you only write 1% of the best seed checks. If you don’t eat the pie, someone else will. If someone else eats more pie than you, they will be stronger tha
700.
by theptip 2y ago
The advantage of your position is that being a tech lead gives you a lot of ownership and responsibility, and that is one of the best ways to learn a lot in a short time. The disadvantage is that it’s hard to grow beyond what you can visual
701.
by theptip 2y ago
In all the systems I’ve built (mostly Django) you need to tolerate vN and vN+1 simultaneously; you are not going to turn off your app to upgrade the DB. You’ll have some Pods on the old application version while you do your gradual upgrade.
702.
by theptip 2y ago
You’re missing the point being made in this thread, which is that there might be subtle long-term impairments to the genetic condition described in the OP. Indeed the article discusses this thoroughly, noting that since it’s a very small sa
703.
OpenAI Email Archives(lesswrong.com)
48 points by theptip 2y ago|6 comments
704.
by theptip 2y ago
This is the simple solution. 10 years seems fine to me, 20 years at a stretch. The current regime is a clear case of regulatory capture.
705.
by theptip 2y ago
This is the American model of anti-trust, very much not the European model (which explicitly targets competition for its own sake even when consumers are not harmed by the monopolistic behavior).
706.
Crisp and Fuzzy Tasks [in AI](aligned.substack.com)
1 points by theptip 2y ago|0 comments
707.
by theptip 2y ago
This is a crazy goal-post move. TFA is proving a positive capability, and rejecting the null hypothesis that “LLMs can’t think they just regurgitate”. Making some illegal moves doesn’t invalidate the demonstrated situational logic intellige
708.
by theptip 2y ago
When I say “pay to create” I generally mean authoring new material, distilling your career’s expertise. Not my field of expertise but there seem to be experts founding startups etc in the ASIC space, and Bitcoin miners were designed and bui
709.
by theptip 2y ago
That’s what your VC investment would be buying; the model of “pay experts to create a private training set for fine tuning” is an obvious new business model that is probably under-appreciated. If that’s the biggest gap, then YC is correct t
710.
by theptip 2y ago
I disagree with most of the reasoning here, and think this post misunderstands the opportunity and economic reasoning at play here. > If Gary Tan and YC believe that LLMs will be able to design chips 100x better than humans currently can
711.
by theptip 2y ago
> As a nurse, you end up relying on intuition a lot. It’s in the way a patient says something, or just a feeling you get from how they look There is a longstanding tension between those who believe human intuition is trustworthy, and the
712.
by theptip 2y ago
> “I think it ultimately does fall within the line of being ethical, but it isn’t a slam-dunk case” I concede that I haven’t thought as deeply about this as ethicists, but I strongly suspect that the cost/benefit calculation here is
713.
by theptip 2y ago
How many have you founded? Important data point for such an opinion.
714.
by theptip 2y ago
> Combining two broken systems - compromised search engines and unreliable LLMs - seems unlikely to yield that vision Counterpoint: with a chain-of-thought process running atop search, you can potentially avoid much of the meta-search &#
715.
by theptip 2y ago
The goal of these cost cutting initiatives is not an absolute reduction in cost, but a relative one. They needed to show an improvement in operating margin, ie % of revenue spent on engineers. If your engineers become 20% more efficient the
716.
by theptip 2y ago
The answer depends on what is rate-limiting growth; while we are supply-constrained on GPUs you can’t just increase AI usage. The next bottleneck will be datacenter power interconnects, but in that scenario as you say you can expect power u
717.
by theptip 2y ago
> You think you're being smart, you're just being another annoying Ackshually guy We are grown ups, we can disagree without resorting to ad homenim. (Might be time for you to review the HN code of conduct.)
718.
by theptip 2y ago
Don’t give them more ideas!
719.
by theptip 2y ago
Great example. I do think it’s a grey area to knowingly cause some potentially untrustworthy site to be loaded as the OP did (even if it’s a white hat domain now, that might not always be true). .gov should offer these detection services, a
720.
by theptip 2y ago
It’s a shame they learned the exact opposite lesson from what they should have. In fact they should have added their own honeypot company names to the DB to force companies to parse robustly.

More ›