aleqs

What bizarre and absurd line of reasoning. Users who care about their privacy and opt out of downloading ads and malware are 'denying creators revenue'?

Are you denying creators revenue by not reading reading/observing every ad that comes your way and making purchases based on them? Maybe you should read/comment on HN less and focus on consuming more ads instead?

What at an incredibly stupid thing to say.

> over-arching mission of making the web MORE accessible, not less

Right, that's why they pushed AMP and upranked AMP pages in their results. That's also why they decided to severely neuter/remove as blocking extensions for Chrome. That's also probably why google search results are getting worse by the month with more and more ads and spam being upranked to the top.

It's because google has a mission of making the web more accessible. Okay bud.

I think companies will self host (including on rented hardware) even if it's more expensive, and that, along with efficiency improvements, will drop demand for big AI. I think big AI is overspending on hardware/datacenters at the moment.

Very much agree - efficiency improvements are very real both on model and hardware side. The reliance on proprietary OpenAI/anthropic APIs is a problem though, one that will naturally resolve itself in the favour of self-hosted/open models.

Hard agree - the benefits of local/self-hosted models are not just hardware/cost (it might be more expensive at the moment), but what you get in exchange is unnerfed/unstupified models, full cost/usage transparency, optimized/specialized models, privacy/security, etc.

I think hardware prices will come back down once we start seeing more efficiency improvements in models and hardware, and once more people and companies self-host models (which seems to be happening more and more these days). I think the massive infra/hardware expenditures of OpenAI and the like are going to end up unnecessary, leading to hardware price drops.

Ah yes, agreed!

Absolutely love this!

While I agree with everything you say, the reality of most tech companies these days (based on my experience), is that they will not even invest time into maintaining their own infrastructure and libraries unless forced to do so - much less OSS. Building useless features for gaming metrics, enshitification, dark patterns, borderline malware/hype integration - all would be prioritized over foundational infra/library investments.

It happens, but I know it's not enough, and I semi-regularly dream about my projects and AI coding at this point. I do take breaks as well, for food/showers/etc, so not as crazy as it sounds - though I try to line up any breaks with periods when the bots are working on something long/hard..

Somewhat off-topic but related - doing a lot of AI-assisted coding lately, multiple projects at the same time. I'm basically in 15 hour loops of reviewing code/output and creating prompts, switching to the next project while Claude (and others) work. I noticed an almost dyslexia emerging where I'm thinking of the prompt/instructions while typing, then I look at what I typed and it's not what I was thinking - sometimes it's a combination of two different thoughts/prompts that are actually intended for separate projects. It's so weird - I can instantly recognize it's wrong when re-reading the prompt and I still have my intended prompt/instructions in my head. I've never been diagnosed with dyslexia or ever had similar things happen, before AI seemingly captured my brain with the promise of delivering one some of the dozens of my dormant projects/ideas. Maybe I need a break...

Thank you!

yeah similar concept for repo structure/hygiene - I wasn't aware of vale before but cool project!

I'm working on a general repo shape/structure linter (language agnostic)[0] - the idea is to enforce things like directory structure, existence of various files (LICENCE, etc.), file naming patterns, jsonpath + schema over json/yaml/toml, absence of potentially malicious unicode. It comes with rule bundles for various languages/presets which can be combined and extended. A goal is for it to be very fast, and useable on huge monorepos. I noticed myself having to add various forms of validation/scripts when coding using AI, and decided to build a reusable, fast tool for this purpose instead of rolling validation scripts for each project.

[0] https://github.com/asamarts/alint

Indeed, I've been experimenting with agent workflows, for complicated tasks - where I essentially have a graph of agents with different roles/capabilities, including such things as breaking down complex tasks into simpler ones. There seems to be a point where a complex enough task is better performed by a group of cheaper agents/models than by one agent using one of the SOTA big models, in terms of both quality and cost.

Is it expressed as an API though? Does Stripe Projects have a documented public API? Also isn't the payments for this involve manually entering payment/card info (meaning it could theoretically be done a dozen different ways which do not require Stripe)?

In theory, this is a cool idea, but in practice I think this being done through a proprietary, locked-in Stripe product, is going to ultimately hinder adoption.

The security implications here are also concerning - from what I can tell - Stripe seems to have access to all of the keys/credentials for third party accounts/resources provisioned via Stripe Projects.

So stripe has centralized control over payments, KYC, credentials/keys (full lifecycle, not just storage), the provider marketplace, and even over the availability/reliability of anything built with on top of Stripe Projects (since now your credential/key lifecycle has Stripe on the hot path).

This is like a more janky/less reliable loveable, without the handrails, and with a mere illusion of less lock-in.

Imo, this kind of thing will only work long-term as an open protocol without Stripe lock-in, and I know certain people/companies are already working in this direction.

What KYC does Stripe Projects do other than email verification?

The problem is, in the minds of these people 'firing at 100% all the time' generally means doing busywork and/or thinking of ways to cheat/manipulate their customers and the market for maximum gain whole delivering minimum value. I would have loved to be 100% engaged working on solving real problems in honest ways at some of my past jobs, but alas MBA/marketing leadership, which has taken over much of tech has very little interest in actually building good things and solving real problems in honest ways.

I think the problem is even more general than that, and has existed since before LLMs. All of the decision makers are incentivized to chase short term gains and ignore everything else. Many tech companies already had huge gaps in knowledge around their own codebases simply because such knowledge and expertise is basically treated as a liability/expense rather than an asset.

I'm actually very optimistic about LLMs/AI for basically the opposite reason tech leadership/MBAs are - I think it will allow us to overcome organizational/business/marketing ing hurdles that tech companies rely on short-sighted MBA-style 'leadership' for in the first place. And not because I believe in OpenAI and Anthropic - I think the future is self-hosted or community -hosted open models, and open collaboration among willing peers, building open software to solve real problems in honest ways, rather than hierarchical top-down corporate hellholes pumping out pre-enshitified crapware full of ads, tracking and dark patterns.

I think intelligence per compute will go up significantly in the coming years, while the cost per compute will drop significantly. No way to know for sure, so I guess we'll see

Indeed, I feel like we are in the early computer equivalent phase of AI, where giant expensive hardware is still required for frontier models. In 5 years I bet there will be fully open models we'll be able to run on a few $1000 of consumer hardware with equivalent performance to opus 4.7/4.6.

The usage metering is just so incredibly inconsistent, sometimes 4 parallel Opus sessions for 3 hours straight on max effort only uses up 70% of a session, other times 20 mins / 3 prompts in one session completely maxes it out. (Max x20 plan) Is this just a bug on anthropic side or is the usage metering just completely opaque and arbitrary?