theptip

24 ms·

451.
by theptip 1y ago
Huh, apparently he is a “white supremacist” for posts like this? https://world.hey.com/dhh/the-waning-days-of-dei-s-dominance... I missed all this drama, it does seem like there is an echo chamber forming over on Blues
452.
by theptip 1y ago
I agree the market dynamics are weird now, I disagree that says much about the existence of other equilibria. For example, inference on older GPUs is actually more profitable than bleeding-edge right now; the shops that are selling hosted i
453.
by theptip 1y ago
> bursty loads that seem typical of AI data centers Don’t datacenters want to run at their rated capacity 24/7?
454.
by theptip 1y ago
OpenAI is profitable if they stop training their next generation models. Their unit economics are extremely favorable. I do buy that they are extremely over-valued if they have to slow down on model training. For cloud providers, the analys
455.
by theptip 1y ago
I wonder if we now have the tools to build unit tests for docs now; an LLM should be able to take on the persona of a beginner try to follow your doc. For bonus points use a dumber/older model that can’t have trained on your API.
456.
by theptip 1y ago
What do you think about him?
457.
by theptip 1y ago
Agreed, but I do think it’s important to distinguish: - voted Trump because they believe the constitution protects us from his worst impulses; would support constitution over Trump - voted Trump and would shred the constitution if they had
458.
by theptip 1y ago
If the president shreds the constitution, there would likely be many in the military opposed to it. While they are actively replacing cabinet positions with loyal outsiders that have little-to-no experience within the organizations they now
459.
by theptip 1y ago
I agree this is a problem. The best reason I can come up with is “because it is fun/beautiful/interesting” (to some people). But IMO that pushes you to making the curriculum more flexible and not forcing every student into the sam
460.
by theptip 1y ago
I believe this is going to be an increasingly important factor. Call it the “shoelace fallacy”: Alice is supposedly much smarter but Bob can tie his shoelaces just as well. The choice of eval, prompt scaffolding, etc. all dramatically impac
461.
by theptip 1y ago
To some extent, but I think it’s easier to have fully automated buffering if you physically control the cars. Eg you can have backup vehicles parked indefinitely nearby if you want, whereas there will always be some unpredictable churn from
462.
by theptip 1y ago
The setup at SFO is currently quite annoying (Lyft/Uber require you to walk 5 mins to the garage roof, and drivers need to park/wait 5-10 mins away, so there's always a substantial delay). Taxis get the privileged parking spo
463.
by theptip 1y ago
Unfortunately, a bit of a one-sided article. The most convincing argument for buying housing is that it hedges you from potential large future cost of living increases. If you can’t afford repeated rent increases then the consequences will
464.
by theptip 1y ago
Sounds like “loss aversion”, which was studied by Kahneman and Tversky. https://en.m.wikipedia.org/wiki/Loss_aversion
465.
by theptip 1y ago
I get your point and would be much more inclined to agree if we were talking about trying to hit a 3x risk increase. But we are talking about huge risk margins here, many OOMs. My problem with your argument is that as framed it’s a fully ge
466.
by theptip 1y ago
Agreed that lumpiness is an issue and so in practice you wouldn’t want to argue for coal levels of death-per-MWh. This concern is, I believe, the crux of why folks are overly-conservative - the few well-known disasters are terrifying and th
467.
by theptip 1y ago
True, but we have multiple OOMs to play with. How about we try to go from 0.03 to 0.3 deaths per TWh and see how much cheaper we can make it? As long as we stay lower than 30 we didn’t actually make a mistake.
468.
by theptip 1y ago
I don’t follow. If nuclear initially costs more than coal, then the first effect as it decreases is displacement when the prices cross over. Then if it falls further you will notice consumer price drops.
469.
by theptip 1y ago
“All of the above” seems a good approach. If this is an existential crisis, why would we not hedge our bets? (Not everywhere has good sun for solar.)
470.
by theptip 1y ago
Sure, but one comes first.
471.
by theptip 1y ago
Agreed, you are talking about non-safety factors. I don’t think they necessitate the price levels we see; for example, look at how cheaply China can build reactors. I think it’s quite clear that we pay a high safety / regulatory premiu
472.
by theptip 1y ago
The goal of making nuclear cheaper isn’t to lower consumer costs. It’s to displace CO2 emitting baseload sources like coal and gas.
473.
by theptip 1y ago
It really is. Nuclear is 100-1000x safer than coal. By insisting on such an aggressive safety target, we force prices up and actually incur much higher levels of mortality - just delivered in the boring old ways of pollution and climate-dri
474.
by theptip 1y ago
I’m already throughput-capped on my output via Claude. If you gave me 10x the token/s I’d ship at least twice as much value (at good-enough for the business quality, to be clear). There are plenty of usecases where the models are not s
475.
by theptip 1y ago
> it is a thoroughly settled matter that you cannot legislate away mathematics. I don’t think this protects us. I view the “encryption is maths” position as referring to backdoor keys. But this time they figured out client-side mandated
476.
by theptip 1y ago
OpenAI and A16Z are vigorously opposing this one. So this should at least discount the simplest anti-competitive conspiracy scenarios.
477.
by theptip 1y ago
> Coding is formal I just don’t see it like this; code is craft, and there are ten ways to solve any given problem. Reasonable people can select different tradeoffs, and taste is also a big factor. Maybe if you are working in very low-le
478.
by theptip 1y ago
A good case study. I have found these two to be good categories of win: > Use these tools as a massive force multiplier of your own skills. Claude definitely makes me more productive in frameworks I know well, where I can scan and patter
479.
AI robots can carve stone statues. buildings are next(fastcompany.com)
11 points by theptip 1y ago|10 comments
480.
by theptip 1y ago
I agree to some extent, but there is a slippery slope to “no rules apply to the rich”. I do agree that in the case of victimless crimes, having some ability to recompensate for damages instead of outright ban the thing, means that we can en

More ›