Lots of stuff involving cybersecurity or reverse engineering for one. At one point Gemini wouldn’t tell me the command line to disable approval on codex because YOLO mode is apparently a hacking technique. I’ve been using Grok as a backend for some experimental AI game storytelling, and obviously fiction isn’t all rainbows and unicorns, so often ChatGPT/Gemini bail out when the going gets rough.
Other things… well people are going to disagree on what’s “legitimate”, but perfectly legal queries about morally questionable topics are commonly refused. ChatGPT refuses to translate or OCR texts containing racial slurs, won’t help with information about building firearms, won’t give you information about how to stage a bank robbery, won’t talk about anything involving suicide, refuses questions around sex work and kink activities.
I can go to the library and rent a book on gunsmithing, sex work is legal, heist fiction is popular, racial slurs are unpleasant but common in certain texts. I totally understand why they want to block that stuff off, but I also don’t really like my computer telling me what is acceptable for me to think about based on someone else’s moral code. Also a world where games don’t include anything immoral or “unsafe” is probably quite bland.
I concur. Generally I find Gemini answers to be the least biased and most factually accurate, without too many of the annoying AI writing quirks.
However, I find the Gemini web app to be by far the worst, and Gemini itself second only to Claude in terms of refusing legitimate requests. It used to be the worst for that, but Claude has really put up the guardrails since their run in with the US president.
Grok has no concept of safety, which means that it can do certain things that none of the other models are allowed do, especially when it comes to research, creative tasks, humour and games.
> PE companies are not especially interested in long timelines, whereas companies can eventually provide a lot more value that they’re worth right now.
The value of a company does include the value of future returns. The standard model is net present value. In theory, if the market is valuing those future returns correctly, a PE fund would probably just do the value maximisation through investment and genuine business rationalisation, and skip the financial engineering phase of the cycle. I believe this is actually what most PE managers would prefer to do. They generally don’t actually want to burn their efforts for a short term fake valuation boost. They are still people after all. It just seems to be what the market wants them to do, and they prefer money.
> the market is pretty bad at valuing companies
The market has always been bad, but probably more randomly bad, historically. Different people had their own finger-in-the-air methodologies and an estimated market value of a company had a lot more random noise around it.
The issue is now that ~every large institutional investor is valuing companies in the same wrong way, which creates opportunities for, essentially, an arbitrage between reality and their dumb models which these types of PE funds are exploiting.
It also creates systemic risk to the financial system. When everyone is making the same mistake, “independent” market participants aren’t really independent. This is in essence what any bubble is, but usually isolated to people misreading the fundamentals of a particular sector. If the techniques behind financialisation are themselves a bubble, however, we could be in for one hell of a pop if the market realises. Like when the market realised they’d been valuing subprime mortgage books wrong in 2008.
I have no issue with individuals choosing to listen to generative AI. I even occasionally listen to it myself when I’m deep working and just need to occupy that part of my brain (having previously listened to algorithmically generated music or those endless copyright free trance mixes for the same purpose). But I don’t like how it’s flooding discovery platforms to the point that it gets impossible to wade through slop and find actual bands that I could see in person.
It’s like when Etsy turned into a Made in China marketplace. MIC is fine, but if I’m going to Etsy it’s because I wanted something else.
Bringing morality into it opens a whole can of worms that I don't think we have the tools to answer.
My view is companies don't have a conscience, and any expectation that they are going to independently act with moral righteousness is unrealistic. Any perceived conscience is either for marketing (green/pinkwashing), or the sum of the morals of their owners multiplied by their willingness to exert any moral authority over the company.
Besides, if you try to imagine a company having an independent conscience, what even would that conscience be based in? I'm vegetarian and think it's immoral to eat meat, but obviously I'd be insane to expect companies to divest from meat based on my peculiar moral position.
In most cases, people do not exert any moral authority over anything they own. Do you actively select your pension investments based on your morality and vote in the shareholder meetings? If you do, I'm genuinely pleased and happy that someone is. But the reality is most people don't give their investments any thought beyond "line goes up", so companies end up acting as ROI maximisers.
So: the main way we enforce morality on companies is ultimately the government. If you want companies to act morally, you set the rules such that an ROI depends on following our democratically agreed set of regulations. Maybe that even harms economic growth but we still consider it worth it (which is typically how we think in Europe, but look at our economies are doing!). However, the company and its investors are still acting as ROI maximisers.
That's only possible if the financial system is valuing things systematically incorrectly.
IFF a company is truly, honest to god, less valuable than the sum of its parts, then it (or the subset that would have more value to someone else) SHOULD be dismantled, and those resources sold and reallocated to more productive use. You probably make these sorts of decisions in the capacity of your own personal finances without even thinking about it.
On the other hand (and what I believe is likely happening is) if cynical financial engineering is allowing you to turn a useful company that's valued poorly by the market into a useless company that's is paradoxically highly valued by the market, in the short term, and that keeps happening over and over again, then the tools used to calculate the market value are wrong.
This is illustrated by how PE commonly trashes trusted brands. A brand doesn't show up in your EBITDA. If you trash a brand quickly enough by cutting costs and quality, some institutional sucker will buy the company because they haven't clocked that the current EBITDA is elevated due to asymmetry in how quickly the costs come off and how quickly the revenue falls off after burning the brand.
This is just the design of a PE fund. They run on a fixed cycle, so early on they heavily invest into their portfolio with the aim of resolving that risk and maximising the sale value by the end of the cycle.
In principle, I don't think there's anything wrong with this. All investment expects a ROI over some time horizon. Public companies do the same thing. Anyone who founds a start-up is doing it too. The only real distinguishing feature of PE is how successful they have become at aggressively optimising for market value.
The issue is that the sale value at the end of the cycle can be massively influenced by cynical financial engineering. This seems to me to be more of an issue with how every institutional investor apparently now prices companies purely on reductive metrics like EBITDA x the industry standard multiple.
The cause of the rot is widespread over-confidence in dumb financialization models shaping the system.
(Or, since it's HN: if your machine learning model is training well, but misaligned with real life: do you blame AdamW?)
Is there any proprietary Amazon end-dev/ops facing service that's worth using? I've never had a good experience with any I've tried - CodeBuild, Cloud9, Q, SageMaker, WorkMail, WorkDocs, Chime, OpsWorks,...
I love AWS at the infrastructure level, but their PaaS tends to be meh, and their end-user directed stuff is usually atrocious.
Microsoft have historically tended to dogfood their own products.
Obviously you want to be aware of what else is on the market, and use the right tool for the job -- but equally if you have a directly competing product, you'd prefer your org's telemetry and suggestions are directed towards improving your own software rather than your competitors'.
I'm sure there's some bean-counter calculus involving recruitment, PR, demonstration of capabilities, they were going to be doing training flights anyway so why not do a few in public, etc. but they're more rationalisations rather than reasons.
I hope it stays that way too. A world where we take everything away unless it fits into the 5 year ROI spreadsheet sounds dreadful. In any case there'll a long tail of nth-order outcomes that we can't simply reduce down to a risk-reward calculation.
There's probably some deep reason why humans just have a drive to show off their awesome stuff.
People oppose data centres mainly for local reasons.
We already have more than 10k Starlink satellites, and there’s almost no outcry about that outside of astronomers, who are justified in their concerns… but politically irrelevant.
Plus, you can technically launch a satellite from almost anywhere if you’re not picky about the orbit, so you just need to find a single country willing to give you a logistically viable launch site. There are no international laws that would prevent, e.g., India launching 1 million satellites.
I live in inner London within the ULEZ zone. ULEZ is about local pollution not GHG and even then the newest cars it affected were 8 years old. My 15 year old car was fine. If anything it didn’t go far enough.
I have an ICE and don't feel like I'm being punished? It's exactly the same as it's always been. If anything, we're still benefitting from ridiculous subsidies and driving an ICE should be a lot more expensive than it is.
The reason for the EV nudging is it's a chicken and egg problem. The government doesn't want to run a national charging network themselves for obvious reasons, but private investors don't want to build it out either until it can make money.
So they've been trying to fix it from both sides, both by incentivising EV ownership and encouraging EV charging infrastructure. They're also trying to make charging at home easier, even if you don't have a driveway, by installing those little channels you can run the cable through.
Yes, the government are putting their finger on the scale in favour of EVs. Nobody's pretending they aren't. If combustion taxes were as high as they need to be to account for the externalities, the economy would collapse, but we need to get off ICEs for myriad reasons. Seems like they're doing a pretty good job overall and the main problem is just our high electricity price.
You're paying the au pair partly in accommodation, food, bills and a visa. The visa isn't coming out of your bank account, but it's definitely part of the incentive, so you could see it as a government subsidy.
For comparison, a full time "virtual assistant" with fluent English from the Philippines costs upwards of $700/month nowadays.