Hacker News new | past | comments | ask | show | jobs | submit | suddenlybananas's comments login

I find it hard to see the advantage of scraping reviews versus just looking at the actual score.

Score is rigged, description does not match reality, negative reviews are deleted.

AI is perfect to answer questions like:

- is there a big table on pictures (proper workspace)

- is there a direct sea view

- any mentions of drain issues...


Only if you know what to ask.

You'll miss all the things you didn't ask about -- good and bad.


I find that the only thing that works with Airnbnb is just to read all the reviews after excluding anything that isn't close to a 5.0 rating. Some kinds of negative comments in reviews can safely be ignored (e.g. the people who rented an apartment in the middle of downtown and were surprised that it's noisy), but almost any substantial negative comment in any review should be a red flag, no matter how many positive reviews there are.

for both Booking and AirBnB make sure to read recent reviews. Booking especially likes to default to "relevant" reviews (which are always glowing reviews from five years ago)

It's remarkable how uncomfortable AI guys get about criticism. Why is it a bad thing to criticise the models? The models still have enormous gaps in their capacities (e.g. river crossing questions where you make them simple instead of difficult).

Well it's true that all of the most recent advances come from changes the architecture to do to inference scaling instead of model scaling. Scaling laws as people talked about in them in 2022 (that you take a base LLM and make it bigger) are dead.

I think you want both. To scale the model, e.g. train it with more and more data, you also need to scale your inference step. Otherwise, it just takes too long and it's too costly, no?

Bit weird to rename something after 400 years for purely nationalistic reasons though. Maybe Mexicans will be up for it though if Americans accept being called Unitedstatesians in English.

The name is now finally NOT centered around one nation.

Plus, it's the opposite of weird. National reasons are the most frequent cause for renaming areas after hundreds of years.

Additionally, the order doesn't change the name of any gulf along Mexico's Eastern shoreline. So whether they're "up for it" is inconsequential.


>The name is now finally NOT centered around one nation.

'Gulf of the Americas' would make more sense in that case. But that doesn't project the intended message from the new administration.

You can justify it however you want, but the intention was not to be inclusive and I think that's pretty clear. Unless talking about annexing Greenland and absorbing Canada are also just ways of making us one big happy family, I think the intention of the name is clear, regardless of how much sense one can force it to make after the fact.


The name predates Mexico by 200 years. The name refers to the city of Mexico, smack dab in the middle of the gulf.

Salient point, the US is the only American country with America in it's actual name. Names of things change. It's how language works. Even though I think Gulf of America is actually more apt a name, I and most people in the US could care less what it's called. It's just Trump playing power games like China does with the "South China Sea" v. "Sea of Japan".

Just call them USians. When I say 'americans' I often talk about north Americans, I use USians to disambiguate.

I have never heard American refer to anything but a United States resident, so I seriously doubt you. USian is disrespectful to the preferred demonym, if you do care about that.

Pedantically, the formal name of Mexico is Estados Unidos Mexicanos, or United States (of) Mexico. USians would then be ambiguous.

I think that it should be run as a public service like utilities and should be as cheap as humanly possible. Why not?

I personally like the notion of a common public infrastructure that subleases access. We already sort of do that with mobile carriers where the big 3 provide all access and all the other "carriers" (like google fi) are simply leasing access.

Make it easy for a new wireless company to spawn while maintaining the infrastructure everyone needs.


My public utility is bad at its job because it has literally zero incentive to be cheap, and thus my utilities are expensive

> it has literally zero incentive to be cheap

Do private utilities have any incentive to be cheap?

The reason we have utility regulations in the first place is because utilities are natural monopolies with literally zero incentive to be cheap. On the contrary, they are highly incentivized to push up prices as much as possible because they have their customers over a barrel.


Utilities do have incentive to be cheap as long as there are is the presence of competing offerings and the lack of collusion.

...which is unusual with many utilities, but is also pretty common with wireless carriers in much of the world.


I believe the idea is that you shouldn't have a corporation provide the utility if there's only going to be one.

"public utility" implies it's owned by the public not a profit seeking group of shareholders.


A private electric grid is a nightmare. Look at Texas. People pay more, and they get less coverage. It's worse by every metric. The conversation should revolve around, how can we fix the government so that it isn't 5 corporations in a trench coat who systematically defund public utilities and social safety nets in hopes of breaking it so they can privatize it and make billions sucking up tax payer money while doing no work. See the billions in tax funding to ATT, Google, etc... to put in fiber internet that they just pocketed the cash and did nothing.

In Texas electricity is literally less than half the price than the price in my state on average. (14c/kwh vs 34c/kwh) (I live in California)

If you want to say its worse, perhaps you should check if its actually worse first.


A big part of the reason that California average electrical price per kWh is high is that a huge portion of the cost is fixed costs, and California's efficiency push has resulted in the lowest lowest per capita electricity usage (and fourth lowest per capita energy usage) in the USA, so the fixed costs are spread over fewer kWh.

Conversely, Texas has significantly above average use per capita, spreading the fixed costs across more kWh, but still results in higher annual costs per capita, despite lower per kWh rates.


Let's also not forget the cost of the things like the campfire fire. That's a huge bill that needs to be paid and that cost is ultimately going to come out of the kwh rates.

Further, the LA fires might have also been caused by a downed line so that's going to be a fairly big cost to the power company.


> Let's also not forget the cost of the things like the campfire fire.

That's, I assume, a reference to the 2018 Camp Fire.

> That's a huge bill that needs to be paid and that cost is ultimately going to come out of the kwh rates.

The Trust established to pay PG&E liabilities for the 2015 Butte, 2017 North Bay, and 2018 Camp Fires, which discharged PG&E's responsibility for them, receives no additional ratepayer funds after its initial funding and is in the wind-down process expecting a single final top-off payment to already approved claimants. So, no, its not a huge bill that will be paid out of future rates.


Because competition drives innovation. 5G exists as widely as it does because carriers were driven to meet the standard and provide faster service to their customers.

This article is essentially arguing innovation is dead in this space and there is no need for bandwidth-related improvements. At the same time, there is no 5G provider without a high-speed cap or throttling for hot spots. What would happen if enough people switched to 5G boxes over cable? Maybe T-Mobile can compete with Comcast?


Well, 5G is unlikely to be built in my area for the next decade, meanwhile 3 operators are building networks in the slightly more populated areas.

Coverage requirements could be part of the spectrum auction, like when Google managed to get "no SIM locking" part of the spectrum requirements for the 700 MHz band, opening up Verizon phones https://en.wikipedia.org/wiki/2008_United_States_wireless_sp...

Competition drives innovation, but also, we've generally seen that things like municipal broadband are _more_ innovative than an incumbent monopoly carrier. Large chunks of the US don't have much competition at all in wired services, and if we approach that in wireless, we are likely to see the same effects starting where the local monopoly tries to extract maximum dollars out of an aging infrastructure. Lookin' at you, Comcast, lookin' at you.

As you say, "incumbent monopoly carrier" is not competition, so a municipal provider which competes with broadband is a great idea. This article, however, is arguing we don't need more bandwidth, and we need more consolidation of major providers: I'm not convinced.

The T-Mobile 5G Rely fixed-wireless home internet plan offers no caps and no throttling plans.

It does past a terabyte.

The fine print does say:

> During congestion, customers on this plan may notice speeds lower than other customers and further reduction if using >1.2TB/mo., due to data prioritization

So not really a cap, but a deprioritization. A few friends using it around me routinely use >2TB/mo and haven't experienced degradation, I guess there's not excessive congestion. YMMV.


Three things are necessary then:

1. It must be well-run.

2. It must be guaranteed to continue to be well-run.

3. If someone can do it better, they must be allowed to do so - and then their improvements have to be folded into the network somehow if there is to be only one network.


Internet is treated this way in Germany, and it's slow and expensive. Eastern European countries that put their bets on competition instead of regulation have more bang for the buck in their network infrastructure

How come it's failed to provide cost effective internet in the US then?

Do you believe in out-of-india theory for IE or are you just sceptical about the use of the Rig Veda specifically.

It gets a bit silly when you start using archaeology to prop up modern political doctrines. Humans left Africa over 100k years ago, and groups have been moving around ever since. Whether a group moved from the Caucasus to South Asia or vice versa, around 5k years ago, shouldn't really matter. Perhaps they had moved in the opposite direction 10k years ago. Obviously, we all have human ancestors who were living 5k, 10k, 100k, 200k years ago.

> Whether a group moved from the Caucasus to South Asia or vice versa, around 5k years ago, shouldn't really matter.

It's not that it "matters" in a political or nationalistic sense. That's an error in interpretation of the motivation for this kind of work.

It is important because the more we know about how we got where we are, the better.

Science is useful, if it is not immediately obvious, then future generations will surely find an use for it, as it has happened time after time with mathematical ideas.

I would even say it is you who are putting a modern political spin on this by rejecting it.


The research is fine. I'm referring to the likes of Hindutva trying to establish that the Aryans were "indigenous" to India and subsequently migrated elsewhere, thus proving that Hindus alone are indigenous to India. The out-of-india theory referred to above.

Hindutva says Islamists and white Europeans were invders. Hows that wrong?

It's correct, but so what? It's part of history, nobody alive was involved, and it shouldn't have political implications for how we want to live today.

Ah, I see. I agree with you then.

Exactly. Humans are moving constantly, even today. It is silly to attach political doctrines to such complex events.

For all we know, we might never get a complete picture and there might be many other aspects which we are not aware of behind PIE.


Option#2: I was only curious about the GPs claim which added Rigveda to the mix.

To what degree can you really trust these data? They obviously have a financial interest in playing with the numbers.

Have you noticed that the numbers are all relative?

Unless they are trying to mislead competitors (who don't look at their own numbers...), they have no reason at all to game those numbers there.


I would say it is trust worthy because if it were found to be gamed then Anthropic’s reputation would crater.

But, we found out that OpenAI is/was gaming benchmarks (https://news.ycombinator.com/item?id=42761648) and that seems to be forgotten history now - so I don’t know.


> I would say it is trust worthy because if it were found to be gamed then Anthropic’s reputation would crater.

But on the other hand, how would we found out that they've gamed the numbers, if they were gamed? Unless you work at Anthropic and have abnormally high ethics/morals, or otherwise private insight into their business, sounds like we wouldn't be able to find out regardless.


I translated from French to English and vice versa and the voice sounded nothing like me in either case. The English to French translation also made me sound about 90 years old.

It depends how much tolerance you have for mistakes. For a waiter or asking directions or things like that, 100% this works great. For a diplomatic discussion where nuance is very important however... It also doesn't work great for translating works of art where the translation itself is open-ended and can be done in a bunch of different ways and requires a lot of editorial/artistic decisions from the translator.

>They came into this with the assumption that LLMs are just a cheap trick. As a result, they deliberately searched for an example of failure, rather than trying to do an honest assessment of generalization capabilities.

And lo and behold, they still found a glaring failure. You can't fault them for not buying into the hype.


But it is still dishonest to declare reasoning LLMs a scam simply because you searched for a failure mode.

If given a few hundred tries, I bet I could find an example where you reason poorly too. Wikipedia has a whole list of common failure modes of human reasoning: https://en.wikipedia.org/wiki/List_of_fallacies


Well, given the success rate is no more than 90% in the best cases. You could probably find a failure in about 10 tries. The only exception is o1-preview. And this is just a simple substitution of parameters.

Join us for AI Startup School this June 16-17 in San Francisco!

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: