I don't know how you think a b2b company could run sales without a CRM like Salesforce.
To give your question a generous interpretation, Salesforce is more valuable than Apptio or your home grown CRM because it already has all the features any sales org needs, and all the fragmented sales and marketing tooling are already integrated with it.
And Sales is a very expensive and also high ROI activity. You don't want your sales team hung up trying to figure out how to get the random CRM to do something. You're not looking to cut costs in this area, you're looking to enhance the overall productivity of the org. Sales tooling overall is very expensive for this reason, any marginal edge is worth a lot.
It's also worth noting that a big value of things like Salesforce is that it lets management check up on what people are doing, because as much as HN doesn't like to admit it, people are often not very careful or diligent, and you need to perform supervision on the vast majority of people to improve their performance.
Jira is similar, in that eng is very expensive, and its probably better than what these companies were doing beforehand, even if it is suboptimal.
It's true, literally no b2b sales companies existed before Salesforce. We must all continue to pay for Salesforce and support its workflows for now until the endless future, lest b2b sales vanish again.
> Especially as the cost of producing code drops, the value of libraries decreases.
Does it? If the cost of slop that (1) no one understands, and (2) no one can be sued for if it misbehaves drops to zero, what have we gained? A "library" is code plus reliability and accountability. (Yes, GPL disclaims liability, but that's why consultants exist.)
I don't think that this is a good idea. For medical applications, I can understand that LLMs are not the best solution, since they are so bad with numbers/probabilities. But for legal advice, I think they should be pretty good.
So the only reason I can think of to forbid such use cases is that people in those professions fear being replaced by machines.
Preventable medical errors kill 250,000 American every year, I can imagine LLMs could be both good and bad for that number, but on net, it is hard to say without just guessing. But if you ban the application of LLMs to medical care, you close that door before even seeing the potential on the other side. I think that is absurd.
I don't think that conclusion really follows because I don't think the ban works that way.
There's a big difference between ChatGPT writing a prescription and a doctor double checking his diagnosis using some kind of Claude code for medicine. ChatGPT writing prescriptions and giving medical device directly to people should absolutely be prohibited for now, but the second approach should be encouraged.
It really isn’t. How many surgeries do you think LLMs perform? How many of those medical errors would’ve been resolved by a chatbot? It’s easy to quote a big scary number and pretend like it has some vague relevance when you don’t actually understand the problem space.
Ok, so how many deaths from medicals errors have been caused by and prevented through the use of LLMs (since you say it isn't hard). Can you enlighten us and not leave us guessing?
I understand many deaths due to medical errors are caused by patients misunderstanding the advice they are given. You are saying you know exactly the net value of LLMs in this problem space?
The point of IP is to encourage the creation of new things.
Not all protections have to be ones that give total control like copyright.
I think it's a mistaken assumption that costs will fall to zero. The low hanging fruit will get picked, and then we'll be doing expensive combined AI/wetlab search for new drugs.
If there is any meaningful headroom we will keep doing expensive things to make progress.
> The point of IP is to encourage the creation of new things.
Then why are corporations allowed to milk successful works for all eternity? Why do we have Disney monopolizing films made half a century ago? Why do we have Nintendo selling people the exact same Mario ROMs from the 80s every single console generation?
They should have like 10 years of copyright so they can turn a profit. Once it expires it's over and the work enters the public domain where it belongs. If they want to keep profiting they should have to keep creating new things. They shouldn't be able to turn shared culture into eternal intellectual property portfolios that they monopolize and then sit on like dragons.
There is always drift between intent and implementation, but to be generous here, Disney is generally making new works with their IP and so is Nintendo.
I am somewhat curious what you think shortening the copyright window would do that's so great for the culture though. We already have more than enough IP slop that's just licensed.
They're just as capable of typing prompts into AI, but what they don't have is good judgement of what good work/code looks like, so what's the point of asking a junior engineer to do something vs asking the LLM directly?
Because a lot of stuff doesn't need to be good it needs to be done.
Nobody is gonna lose money because some script that generates yaml for the build process every hour nested three loops instead of two. Intern, AI, junior dev, junior dev telling an intern how to use AI, doesn't matter. If it works for the week it'll work for the decade. If someone needs to pick it apart and fix something in a year it'll either take no time because they know enough to do it easily or it'll be a good low stakes learning exercise for a junior.
Everyone wants to think their stuff is important but 99.9% of code is low stakes support code either in applications or in infrastructure around them.
I disagree with on-prem being ideal for GPU for most people.
If you're doing regular inference for a product with very flat throughput requirements (and you're doing on-prem already), on-prem GPUs can make a lot of sense.
But if you're doing a lot of training, you have very bursty requirements. And the H100s are specifically for training.
If you can have your H100 fleet <38% utilized across time, you're losing money.
If you have batch throughput you can run on the H100s when you're not training, you're probably closer to being able to wanting on-prem.
But the other thing to keep in mind is that AWS is not the only provider. It is a particularly expensive provider, and you can buy capacity from other neoclouds if you are cost-sensitive.
One thing I will say is: I wish I had had some EM/hiring experience before starting a startup.
My technical skills served me very well in year 1/2, but once we started hiring enough people I could definitely feel my lack of experience.
Maybe big tech EM experience wouldn't have helped me a lot, the context is definitely very different, but at least it would have been some sort of baseline to draw from.
The model does not seem to get told what the cards actually do, it's just trying to guess.
Probably less of a problem for Legacy, but it isn't even told what format it's playing. So I expect it's easy to get the bots to do far better than this by giving them more information.
Let me analyze this opening hand:
*Hand analysis:*
- 3 lands (Riverpyre Verge, Steam Vents, Willowrush Verge) - dual lands which is good
- Monument to Endurance ({3}) - a 3-drop
- Accumulate Wisdom ({1}{U}) - 2-drop, likely a blue advantage engine
- Boomerang Basics ({U}) - 1-drop blue spell
- Gran-Gran ({U}) - 1-drop blue creature (1/2)
This is actually a reasonable hand. I have:
- 3 lands giving me mana access (with color fixing)
- A 1-drop (Gran-Gran or Boomerang Basics to choose from)
- A 2-drop (Accumulate Wisdom)
- A 3-drop (Monument to Endurance)
The curve is solid. I have some interactive blue play with Boomerang Basics and early board presence with Gran-Gran. The mana base supports blue well with all three lands. This is keep-able — while not spectacular, it's a functional opening hand.
Oh, that's a good bug report - historically it was just hallucinating card effects so I made the harness throw the Oracle text for all visible cards into the context, but I bet I forgot to do that for the mulligan decision specifically (it's a weird one). Thanks!
Do you have figures supporting that? Because so far everything I've seen points to current inference subscriptions being wildly unprofitable. It may be a bit dated, but I haven't seen any new reports on unit economics coming close. And forward projections for the data center side, which is the pure play of inference itself, says something like $40B of depreciation per year (assuming they take a full 10 years to depreciate) and maybe $15-20B of revenue to make up for that.
Now, we're still talking about "some analyst" and the most undifferentiated pure play for the underlying economics of inference itself as a whole, but I think that the latter, at least, should remain relevant because if the underlying inference doesn't work on current subscriptions, then nothing built on top of it without significant additional charge will.
There will be consolidation has few players will have the revenues to justify training their own model. Google has enough cash and revenues to be one of the survivors of this race. Openai and Claude will survive in some form or another, at least as a brand. xAi will burn through SpaceX revenues and capital so it will stay around for a while. China will keep subsidizing models. Meta might keep a subpar model around. It still a race for relevance so not everyone will make the cut.
I can go on Gemini, claude, mistral, (and even chatgpt) for free (even through a VPN), so they are definitely not profitable for me as they are not getting anything from me. are you saying subscriptions are subsidizing me?
I don't know how you think a b2b company could run sales without a CRM like Salesforce.
To give your question a generous interpretation, Salesforce is more valuable than Apptio or your home grown CRM because it already has all the features any sales org needs, and all the fragmented sales and marketing tooling are already integrated with it.
And Sales is a very expensive and also high ROI activity. You don't want your sales team hung up trying to figure out how to get the random CRM to do something. You're not looking to cut costs in this area, you're looking to enhance the overall productivity of the org. Sales tooling overall is very expensive for this reason, any marginal edge is worth a lot.
It's also worth noting that a big value of things like Salesforce is that it lets management check up on what people are doing, because as much as HN doesn't like to admit it, people are often not very careful or diligent, and you need to perform supervision on the vast majority of people to improve their performance.
Jira is similar, in that eng is very expensive, and its probably better than what these companies were doing beforehand, even if it is suboptimal.
reply