ianm218

I mean the company has like 3k employees or more right? Lots of them are just working on more applied AI use cases that don't require frontier AI just the right integrations and structure etc.

Opus 4.8/ GPT 5.6 level models with the right workflows/ data/ access are still good enough to do huge amounts of economically valueable work.

What has been the best way to determine return on the AI specific capex for hyperscalers?

I would naively expect Microsoft’s to be the highest since they are probably mainly just selling access to their capex through cloud since they aren’t seriously pursuing frontier AI, I’d imagine Google to be in the middle (selling TPUs, general cloud GPUs, Gemini, revenue lift on ads from better AI) but also spending heavily on infra to compete with OAI/ Anthropic, and then Meta to be on the low end since they are likely getting serious revenue lift from AI but not monetizing their models by API.

> Maybe they figured out it was too easy now, and this is the clamping down we are seeing in response

His post was from the start thinking out loud. It’s a solid contribution. It seems against the spirit of the forum here to respond to someone unauthoritatively thinking out loud with “you’re just making stuff up!”

Very cool! Well let me know if you get a chance to use it would love to hear if it ends up being the right tool.

Heh I was going to queue up a skill that uses the CLI to make it easy to convert scripts to older or newer versions.

If you have any Lua use cases let me know I’m looking for more real world use use cases to justify the effort here.

Hmm I think in general I would not recommend doing this just because Lua does change pretty significant things from 5.1-5.5 so you could have some really hard to understand behavior if some portions are 5.1 vs 5.4 for example.

I think where it would be most helpful is converting a codebase and being able to easily run tests to ensure behavior is the same.

I created a github gist https://gist.github.com/ianm199/5ba0366376eca673142e1f0c79b4... that explains what is practical (I used AI for this to be clear feel free to skim).

Do you have a use case in mind? Would love to chat or take a look at an github issue if you create one.

Performance dashboard is linked here https://ianm199.github.io/omnilua/performance.html.

In a mix of official and unofficial benchmarks wall clock performance is ~1.4x as fast as C Lua and the memory usage is ~1.7x.

So performance is worse to be clear but within range. There’s some performance improvements I haven’t gone for yet that would get it down to ~1.1 I think.

Tangently related but I’ve been deep in Lua recently working on a rust implementation that supports Lua 5.1-5.5 in one Rust Binary https://github.com/ianm199/omnilua.

My ultimate goal was to support LuaJIT in Rust as well but this does not make it easier.

To get more concrete are you using coding agents like Claude Code/ Codex/ opencode etc? What kind of work are you doing specifically?

If you are doing the kind of median enterprise tech work these tools are just good enough to do it at a relatively high level or atleast heavily augment people doing it.

Examples would be like adding routine CRUD features to APIs/ improving observability/ adding tests or accessibility features to codebases etc.

Can you describe your experience using modern AI tools that led to this conclusion? It is hard for me to wrap my head around how my perception could be so different from someone else in presumably the same or similar profession. I'm not asking this in bad faith either but I think your getting downvoted because your comment comes off as a pretty strong assertion without giving details on how you got there.

So is your position that i.e. the Five Eyes [1] cyber security leaders are just pretending that AI cyber security is a serious thing to play into the geopolitical east vs. west thing and its not genuine?

It just feels like people are starting to reach for conspiracy theories rather than engage with the idea that these models might actually be dangerous.

[1]. https://thehill.com/policy/technology/5936339-ai-cybersecuri...

What makes you think that? The rate of change seems to have been increasing and there is so many chip and model best in different directions at the moment.

Rules exist for a reason, it shouldn't take 4 years to get an answer on how to proceed in one direction or another. In NYC for example you often need to hire a "permit expediter" just to get a yes/ no/ maybe answer on your apartment renovation within a year. These bureaucracies are often literally impenetrable without professional help.

> First of all, I don't accept a discussion where humans are placed on the same level as bots or agents. Even if the cost is the same, humans MUST be the choice

Sure but getting more done with less is just how productivity gains work. If your talking about crushing an iOS backlog like the OP is it surely will not cost as much as a human to move along tons of bug fixes and accessibility changes etc.

Here are some case studies on recent anecdotal producitivity gains from AI agents https://cursor.com/blog/amplitude, https://devin.ai/customers/ramp, https://www.warmly.ai/p/blog/claude-code-best-practices.

The third one specifically says that a team of 8 "ships like a team of 24".

> Also, trillion dollars valuations are really justified by random solo-devs creating automatic companies of one that could be companies of four?

I feel like your greatly expanding the scope of this thread, I don't really know how to value these things.

There are atleast some restrictions in localities I.e. https://www.theguardian.com/world/article/2024/aug/12/portof... or parts of Portugal.

Also at a certain point if you ban or heavily restrict energy production to the point that no one can afford to run AC, it is a de facto ban.

This is completely ridiculous you just need to let people use AC properly and it will solve most of the problem with people dying. That doesn’t mean you need to go from place to place in AC or never be out of AC, but every person in southern Europe should have an AC unit in their bedroom and deaths will go down substantially.

Another thing that would help is if Europe stopped being so averse to new construction so that people could build new buildings with central AC that matches the current reality of its climate and demographics, rather than the climate and demographics of last millennium.

This company got valued at $250M https://polsia.com/ and is a one person startup. Lots of people think its more or less a hype job but given "ai agents" have only really existed for a year or two now it's sitll early days.

> Can you say with certainty that this workflow has increased productivity so much that you are seeing profit increases that you wouldn't have otherwise noticed just by hiring more people?

Has the bar raised so much that coding agents need to be even better than any human could do? If the coding agent lets you get done with 1 person what would've required a team of three and you are not going crazy "token maxxing" spending thousands of dollars a week that would be beneficial for a business.

> The reality is the market is contracting and it’s doing that because the extraction and monthly financial models are hurting other businesses which are also contracting because of lower customer demand.

Can you explain what you mean?

> (they need someone who will sell stablecoins in their currency).

It's pretty hard to really lock people out of stable coins really. You really just need someone to sell you some type of cryptocurrency that can be eventually exchanged for stablecoins. You can even do "peer to peer" trades if the government really cracks down on holding crypto.

I agree with the sentiment of this article but atleast some parts of the world with poor currencies like Latam have seen some benefit from stbales.

> Created in 2022, TIP’s mission is to address the decades-old complaint that the agency, traditionally focused on curiosity-driven research with no obvious commercial value, needs to do more to make sure scientific discoveries eventually benefit society—in new jobs and products, improved health care, or a rising standard of living.

This seems like a decent initiative started under Biden’s pro technology/ industrial policy umbrella. It was launched as part of the hugely successful CHIPS act.

Your critique is a false dichotomy. There is a large spectrum between “purely curiosity driven research” and “research that has near term venture outcomes” where applied research lives. Something like the Bell Lans semi conductor and related research would fit into this. Not immediately venture scale focused research, but could lead to it.

Just because Trump is repurposing it for illegal grift doesn’t mean it’s a bad group to put resources into in general.

> Please don't post shallow dismissals, especially of other people's work. A good critical comment teaches us something.

From hacker news guidelines https://news.ycombinator.com/newsguidelines.html

> By levying such a large tax on its other programs, the agency appears to be defying a congressional directive in the final FY 2026 appropriations bill that “No [NSF] directorate shall receive more than a 5 percent reduction relative to the fiscal year 2024 enacted level.” That language was meant to address fears by the research community and some legislators that NSF, if its overall budget remained flat, might decide to grow TIP at the expense of its other directorates—a concern that now appears prescient.

What I find so hard to wrangle is that the Trump admin does almost everything in an illegal hamfisted way, whatever their doing gets stricken down by courts, and then a year later we’re just spending time and resources undoing the obviously illegal things they do.

This change even seems like a positive one I wish they should just pass a bill like a normal government.

I find Metas work very unethical but me but I think they should have basic workers rights still. Like yes what Meta is doing is legal and it’s at will work, but this level of surveillance feels like something the law didn’t really anticipate.

> especially when we all know it’s a veneer and get cold blooded capitalism

This is what I don’t really get about the AI discourse… I don’t have any reason to think this is all a veneer over capitalism. We don’t know what goes on in their head but the track record of the cofounders of Anthropic is quite honest so it feels like a conspiracy theory to suggest they are purposefully getting themselves nationalized when the simplest answer is that there is real documented concern on AI safety risk

> While we're in the token-equivalent of ZIRP, tokens don't cost what they cost, so there's sort of arbitrage to be had

Do you have a source for this? I believe “personal subscription” plans on OpenAI and Anthropic are likely ran at a net loss or close to it, but all indications elsewhere are that API pricing for these companies and likely Google as well are profitable per API call [1][2]. I would definitely believe that the Chinese players are operating at a loss though if that is what you mean.

[1] https://newsletter.semianalysis.com/p/anthropic-growth-and-b...

[2] https://martinalderson.com/posts/are-openai-and-anthropic-re...

If you think the field you are working in is more dangerous and important to the future of national security than i.e. nuclear energy and nuclear bombs you should really not just shut up and act like a cold blooded capitalist.

It feels like people here are not able to separate their personal dislike of the effective altruist types from a real debate on how big the stakes of AI are.

I think your really anchored on anyone successfully breaking restrictions means any restriction is impossible. So your starting from the position that if it is possible for any actor in the world to get past a restriction, then the whole restriction is a farce.

KYC for example does stop most money laundering and financial crime. The most resourced actors like governments/ cartels often find ways around and it is a game of cat and mouse. Normal citizens don't really stand a chance to get around most of them.

Like it feels like your logic is that we shouldn't do background checks for employment because North Korean spy agencies get past them sometimes?

I don't really agree with it but the government is moving towards making you ID yourself to use frontier AI - i.e. only US citizens are going to be able to use Claude Fable supposedly. In that regime the AI companies would in fact know if you are a money laundering expert or a normal software engineer.

> The idea that an LLM can discern intent on any given prompt is farcical.

Not really though. For most people in most situations it's just not going to give you that info. Software security is a niche where its a bit strange in that there is 100X the amount of white hat users than bad actors and there's open source etc.

Isn’t your point that AI safety is impossible to prevent 100% of bad things?

It is quite hard (but not impossible) to get an the frontier AI to tell you how to build a nuke or launder money now, where jailbreaks used to be trivial “ignore all previous instructions”.

It seems like a worthwhile effort.