Hacker Newsnew | past | comments | ask | show | jobs | submit | creativeSlumber's commentslogin

> Owning the decision

Owning a decision means you have something at stake if things go wrong. What would happen to Jack if this decision turns out to be wrong? Any consequences?


Owning stock?

> I've worked at companies that are literally 10x more effective than other competitors in the market purely due to good engineering practices.

Most big tech companies get taken over by leadership with no tech background eventually and the engineering bar drops to the floor.


Are the model weights burned into the silicon / part of the architecture? Or can you update the model weights on these chips? If they cannot be updated, these chips will be outdated the moment they are made given the breakneck speed at which new and improved models are introduced.


https://en.wikipedia.org/wiki/Who_Killed_the_Electric_Car%3F

https://www.youtube.com/watch?v=uep5zOrsAEA

Incredible documentary about the politics around EV's made before Tesla's became mainstream.

People were demanding and protesting asking GM to let time buy their EV'1 after the lease, but they destroyed all the cars. So did Toyota for the EV RAV4's.


I drove a electric Rav4, but couldn't stomach the 40 mile range for the price. But it was a fun car, and it seemed huge (compared to my miata).


> It's all a sliding slope until it reaches a breaking point and falls off like a cliff.

Would it be a gradual decline or a step change after some point?


> A Claude Code subscription should not work with other software.

why not though? aren't you paying for the model usage regardless of the client you use?


No, you are paying to use Claude code… it uses the model underneath, but you aren’t paying for raw model usage. For whatever reason, Anthropic thinks this is the best way to divide up their market.

They want to charge more for direct access to the model.


> No, you are paying to use Claude code

Why would anyone pay a subscription for barebones LLM agent?

You can beat that drum all you want, but you know it's bullshit. People pay the subscription for the AI, not the tool that consumes it. That tool being crap is why everyone started using third-party tools.

The reason they are blocking third-party usage is they want developers to use only their models and no competitors.


Except if you pay the per-token rate, it will work just fine with the other tools. It is only the subscription that doesn’t work.


That's not up to you or me. I think it's pretty clear by the phrase "Claude Code subscription" that it's meant for only "Claude Code". Why are you confused?

This could be so easily abused by companies who spend thousands of dollars per month for API costs you could just reverse engineer it and use the subscription tokens to get that down to a few hundred


That phrase isn't the official one. It's "The Max Plan" which "combines Claude Desktop and mobile apps and Claude Code in one subscription".


Yes, so... pretty clearly not OpenCode.


Yeah, exactly.


It’s like saying Netflix is wrong to require an official Netflix client to access their service. Total dud of an argument if you ask me.


Well, they are wrong, and the argument is still a dud.


Netflix would not even exist if you could just freely download all of the media to your computer and play it anytime because of licensing agreements and other factors. So you can think that they are wrong but that's not really rooted in reality or practicality.


That is possible, was always possible, and will continue to be possible, judging by the availability of 4K rips available via piracy.


Can I script and scrape Claude Code to provide the exact same data for consumption by the banned client? (This sounds like an interesting challenge for Claude Code to try...)


Yes they even offer an SDK for it now so "scraping" is not required


Claude Code provides a headless mode that you can do this exact same thing with:

$ claude -p “fix the eslint in file XYZ”


I don't think they are confused. They are simply challenging the assertion that the model should not work with other software. Which is fair because there is a lot of precedent around whether a service can dictate how it must be consumed. It's not a simple answer and there are good reasons for both sides. Whichever path we take will have wide consequences and shape our future in a very distinct way. So it is an important decision, and ultimately up to us, as a society to influence and guide.


"challenging the assertion that the model should not work with other software"

This has nothing to do with "the model". You can use "the models" through the API for anything.

This has to do with access to a specific product being abused to then get low-cost API access for other use cases


IDK if Anthropic wants to offer a service at below cost, I don't think they should gate keep which client you access that service over. Or in other terms, I won't use a service that locks me into a client I don't like.


How do you draw that conclusion? If Anthropic wants to offer a service at below cost, they seem a lot more justified in restricting how and where they subsidize usage.


Yeah fair, I could've chosen better words. My conclusion is: "I don't want to pay for that", not "they shouldn't be able to do that".


> IDK if Anthropic wants to offer a service at below cost, I don't think they should gate keep which client you access that service over.

Are you going to say why you think they shouldn't? You didn't give a reason.


That seems mutual. They don’t want you to use this service with an arbitrary client and you don’t want to use this service that won’t allow an arbitrary client. So both of you don’t want the relationship. Seems fine.

For my part, I’m fine understanding that bundling allows for discounting and I would prefer to enable that.


> Or in other terms, I won't use a service that locks me into a client I don't like.

Then don't! Or just use the API which doesn't lock you into any client.


But they get telemetry, feedback, good will, etc. That’s one reason why usage is discounted to a subscription fee.


They don't get any telemetry or feedback data from me, as I've turned all that off. So why should I be limited to CC?


If you don’t want to use CC, just use the API


Good will, huh?


You don’t think Claude Code has generated good will for Anthropic? People just liking a brand is powerful.


I don't think this move is generating good will.


A tiny minority being inconvenienced because they are breaking terms of use is fine for them vs pricing claude subscriptions at the true cost.


> aren't you paying for the model usage

No, you’re paying for “Claude Code” usage.


what happens if that large enclosure fails and the CO2 freely flows outside?

That enclosure has a huge volume - area the size of several football fields, and at least 15 stories high. The article says it holds 2k tons of co2, which is ~1,000,000 cubic meters in volume.

CO2 is denser than air will pool closer to the ground, and will suffocate anyone in the area.

See https://en.wikipedia.org/wiki/Lake_Nyos_disaster

Edit: It holds 2k tons, not 20K tons.


CO2 is in general less dangerous than inert gases, because we have a hypercapnic response - it's a very reliable way to induce people to leave the area, quite uncomfortable, and is actually one of the ways used to induce a panic attack in experimental settings.

If it were, say, argon, it would be much more likely to suffocate people, because you don't notice hypoxia the way you do hypercapnia. It can pool in basements and kill everyone who enters.

That being said it is an enormous volume of CO2, so the hypercapnic response in this case may not be sufficient if there's nowhere to flee to, as sadly happened in the Lake Nyos disaster you cited.


CO2 is extremely dangerous in high concentrations because the body reacts and switch off the breathing.


The last section of TFA is called "What happens if the dome is punctured?". The answer: a release of CO2 equal to about 15 transatlantic flights. People have to stand back 70m until it clears.

It would not be good, but it wouldn't be Bhopal. And there are still plenty of factories making pesticides.


Comparing it to X flights maybe correct from a greenhouse emissions standpoint, but extremely misleading from a safety perspective. A jet emits that co2 spread over tens of thousands of miles. The problem here is it all pooled in one location.

Also that statement of 70 meters seem very off, looking at the size of the building. What leads to suffocation is the inability to remove co2 from your body rather than lack of oxygen, and thus can be life threatening even at 4% concentration. It should impact a much much larger area.


It's a gas in an open space, it diffuses very quickly.


Yep. When I had to fill CO2 tanks at a paintball shop yes there were times that I had to open a door (I mean we were talking a lot of fills in short time, btw fills had to start with draining the tank's existing volume so I could zero out the scale) but even indoors a door+fan was enough to keep even the nastiest of sale days OSHA compliant.

Also a 'puncture' is very different from the gasbag mysteriously vanishing from existence; My only other thought is that in cold regions (I saw wisconsin mentioned in the article) CO2 does not diffuse quite as fast and sometimes visibly so...


https://en.wikipedia.org/wiki/Limnic_eruption

I don't know the safety limits for this quantity, I hope the "70 meters" claim was by someone who modelled it carefully rather than a gut check.


Seems like it would depend if there was a small tear or a massive breach.


It also deflates pretty slowly. I'd guess any breeze would remove the hazard altogether.


> People have to stand back 70m until it clears.

How did they calculate that evacuation distance? CO2 is heavy. That little house about 15m from the bubble needs to be acquired.

The topography matters. If the installation is in a valley, a dome rip could make air unbreathable, because the CO2 will settle at the bottom. People have been killed by CO2 fire extinguishing systems. It takes a reasonably high concentration, a few percent, but that can happen. They need alarms and handy oxygen masks.

Installations like this probably will be in valleys, because they will be attached to wind farms. The wind turbines go in the high spots and the energy storage goes in the low spots.


The distance is likely calculated based on the stored volume and the area you cover until the height is significantly below head height (because as you point out CO2 settles to the bottom). Regarding the little house 15m from the bubble, they are not planning to build this in residential areas, so it's very unlikely that there would be a house within 15m just for operational purposes already.


Again, from TFA, the install needs 5 hectares of flat land. I thought it was odd when the article mentioned "flat" land, but I assumed it was more about accommodating the bubble. Now I am thinking it is specifically to avoid the valleys you are describing.


Company says safe limit is 70 meters, about 200 feet.

Easy to build infra and other stuff that far away, especially in locations where this is meant to be used.

There are many aspects of safety

1. If the puncture is due to hurricanes, etc, the danger is non existent. The hurricane will blow away the co2 in no time.

2. If the puncture is due to wear and tear, then the leak will be concentrated and localized. It could naturally diffuse.

CO2 meters can be installed around the unit for indication.

Oxygen masks can be placed around the facility for emergency use.

The dangers are very much mitigatable.


Yeah, I was also immediately thinking about the Lake Nyos disaster. But that one released something like 200k tons of CO2 in an instant, whereas this facility has 2k tons, which would more likely be released more gradually.


So .. significantly less dangerous than a corresponding volume of natural gas, which is also unbreathable but also flammable/explosive?


Why is that a relevant comparison? Is anyone gathering natural gas in giant balloons near habitations or workplaces?



...huh, yeah, I don't love that either. Seems sketchy.

> putting houses around gas holders was discontinued in the UK.


Phased out during my lifetime. I grew up with them as part of the suburban architecture, e.g. https://www.google.com/maps/place/DUNHAM-BUSH+LIMITED/@50.86...


> People will also need to stay back 70 meters or more until the air clears, he says.


Good luck running 70m in a CO2 dense atmosphere. And CO2 hugs the ground it does not float away. It will persist in low areas for quite a while.

Anyone in the local vicinity would need to carry emergency oxygen at all times to be able to get to a safe distance in case of rupture. Otherwise it's a death sentence, and not a particularly pleasant one as CO2 is the signal that triggers the feeling of suffocation.


It's unlikely that the thing will burst and disperse all CO2 immediately. It's just slightly higher pressure than the outside (that's the whole principle). So you have a slow leak of CO2 to the outside. You don't have to run that fast (or run at all).

The way I understood the quote, the safety distance is when they have to do an emergency deflate (e.g. due to wind). The way they calculate the 70 m is probably based on the volume and how large of a area you cover until the height is low enough that you can still breath.

Generally, because it's leaking to the outside, where there is going to be wind it will not stick around for long time I suspect.


> It's unlikely that the thing will burst and disperse all CO2 immediately.

This requires the people running this facility, and all the facilities based on it built by unrelated organizations in the future, to not cut engineering corners on the envelope. I don't take this for granted anymore. But as long as you don't get a big rip, then yeah, it'll be hard to build up a dangerous amount. I wonder if a legally mandatory cut and repair trial on the envelope would reduce risk significantly.

Speaking of wind, I also worry about whoever is downwind if there's a big release. I bet 70m is not quite far enough if it's in the wrong direction.


I wonder whether it'd be possible to augment the CO2 with something that would make it more detectable visually and aromatically, like we do natural gas.

Natural gas is naturally odorless and colorless. Therefore, by default, it can accumulate to dangerous levels without anyone noticing until too late. We make natural gas safer by making stink, and we make it stink by adding trace amounts of "odorizers" like thiophane to it.

I wonder whether we could do something similar for CO2 working fluid this facility uses --- make it visible and/or "smell-able" so that if a leak does happen, it's easier to react immediately and before the threshold of suffocation is reached. Odorizers are also dirt cheap. Natural gas industry goes through tons of the stuff.


I suppose the people working at the plant will be wearing detectors and/or these will be placed at strategic locations in the area.


what do you mean?


It was a joke referring to his essay.

https://en.wikipedia.org/wiki/Bitter_lesson


Not sure why he capitalized bitter...


> Many of the dependencies used names that are known to be “hallucinated” by AI chatbots. Developers frequently query these bots for the names of dependencies they need. LLM developers and researchers have yet to understand the precise cause of hallucinations or how to build models that don’t make mistakes. After discovering hallucinated dependency names, PhantomRaven uses them in the malicious packages downloaded from their site.

I found it very interesting that they used common AI hallucinated package names.


> to prevent wasted processing.

If that was the case, the message should be about a limit on re-enabling the feature n times, not about turning it off.

Also the if they are concerned about processing costs, the default for this should be off, NOT on. The default should for any feature like this that use customers personal data should be OFF for any company that respects their customers privacy.

> You are trying to reach really far out to find a plausible

This behavior tallies up with other things MS have been trying to do recently to gather as much personal data as possible from users to feed their AI efforts.

Their spokes person also avoided answering why they are doing this.

On the other hand, you comment seem to be trying to reach really far trying to find portray this as normal behavior.


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

Search: