lurkshark

Out of curiosity, was this on a codebase where other folks started using Opus 5.5? I’ve been curious if each generation of model has its own biases that make it generate output that’s more difficult for other models/previous models to work with. That’s been my personal suspicion when I’ve hit situations like yours

In the US they call those Prize-Linked Savings Accounts. They’re actually legalized in a majority of states but aren’t marketed particularly heavily (especially when compared to lotteries, sportsbooks, prediction markets, …)

https://en.wikipedia.org/wiki/Prize-linked_savings_account

I don’t think this is aimed at the labs and pre-training, it’s aimed at end users and their agents. Like if you’re a news site the paying customer isn’t a lab scraping your articles for training, it’s an end user that asked their agent to lookup the news of the day

The part I don’t totally understand with the age verification laws is that as I understand it, the websites need to implement the age verification. It seems like the bad actors just won’t do that, and we could’ve made compliance easier for the good actors by just requiring something like the Restricted to Adults label as a meta tag.

https://www.rtalabel.org/

To maybe add some framing, 37% of Americans don’t have enough cash to cover a $400 unexpected expense. Obviously $400 > $65 but I think it puts some perspective on how tight cash flow is for a good chunk of the population.

https://www.federalreserve.gov/consumerscommunities/sheddata...

> it isn't constructed to extract money from you

I mean yes, it is; It’s not a charity. I guess you could argue it tends to do it slower than gambling?

By that logic wouldn’t it be pretty much over for Mac OS as well?

https://gs.statcounter.com/os-market-share

There was a Minnesota company called Healthsense (was acquired by GreatCall which was then acquired by BestBuy, not sure if the company/tech exists anymore) that had a similar approach on a broader scale. Their system used a bunch of mundane smart home sensors in the usual configuration (e.g. contact sensors on doors, motion sensors, etc) but also for tracking patterns and habits, like the refrigerator door, toilet seat, bed, etc. The idea being that an abrupt shift in behavior would trigger a notice for a loved one or nurse to check in. I always thought this was a cool idea and it's a shame it didn't take off a bit more.

The question of "intrusion" was always interesting to me because old folks often face going from nothing to assisted living or nursing home which is often quite intrusive, where somewhat ironically adding a bunch of sensors to your home allows you a bit more privacy.

Kind of a tangent, but I like your type of system as an alternative to the emergency pendants. It always struck me as strange to expect old folks at risk of fall to remember to charge and wear a pendant at all times.

Malus is (well crafted) satire.

They cover a couple reasons on their FAQ

https://atproto.com/guides/faq#why-not-use-activity-pub

It’s a little different because it’s easy for these IPTV pirates to whip up slick branding. Something more like if a guy in a nice looking uniform for a DVD company you hadn’t heard of offered to sell you movies. Especially for folks who aren’t very internet savvy, it can be easy to miss the subtle tells that an offering isn’t legit (even more so when the service works just fine)

I found rpg.actor while looking to see if anybody had done character sheets on ATProto. I really like the work they’ve put into it but they use a static “self” key to make it 1 character per identity/account. That makes a ton of sense for the character-heavy kind of play (like 5e) that it looks like they’re aiming for. I mostly play Cairn and similar games though where characters are a bit more transient. Although they do support DCC which kind of straddles those two worlds

Part of the argument of prediction markets is that it incentivizes good forecasting. Theoretically if you wanted to concoct a novel political polling technique or rent some compute for a new hyper local weather model, you could recoup your costs via the prediction market.

I think in practice the volume of sharp money in the prediction markets is a small fraction and the majority would be better served with the limits you’re proposing

Someone posted this analysis to their Kickstarter comments (they dodged)

https://www.kickstarter.com/projects/tiinyai/tiiny-ai-pocket...

TurnItIn.com was starting to be a thing when I was in high school. I found out it didn’t sanitize the papers you upload and had no CSRF protection, so I could upload a doc with inline JavaScript to hit the change-password and logout APIs.

Was pretty impactful for my education, just not in the intended way

I assume we’ll end up with proof-of-identity attestation as a part of public posting (e.g. Worldcoin) which doesn’t necessarily solve the issue but will at least identify patterns more likely to be LLMs (e.g. a firehose of posts at all hours of the day from one identity). Then we’ll enter the dystopia of mandated real identity on the internet

There are a few “updating” benchmarks out there. I periodically take a look at these two:

https://swe-rebench.com/

https://livebench.ai/

Things like MemGPT/Letta, ToM-SWE, and Voltropy have made long context documentation pretty manageable. You could probably build some specialized tooling/prompts for development artifacts specifically too. But I’ll be the first to admit this is basically “Throw more agents at the problem”

I agree with this, I like spec-driven-development tooling partially for this reason. That being said, what I’ve found is often that I don’t include enough of the “why” in my prompt artifacts. The “what” and “how” are pretty well covered but sometimes I find myself looking back at them thinking “Why did I do this?” I’ve started including it but it does sometimes feel weird because I feel like “Why would the LLM ‘care’ about this story?”

My thought was that if the refusal to service some requests is implemented as an external guard model The Pentagon could try to require them to drop the guard model. This would be similar to saying "we're asking for a 'product' you already 'manufacture'" in the way the DPA is often understood. But if the refusal is baked into the model itself then that argument is dead. Not saying I agree with this, I think it turns into the same kind of problem we saw with the Apple v. FBI conflict and the All Writs Act, but the government doesn't always act in the most sane ways.

I don't think it's that elaborate. I didn't mean to suggest they intentionally goaded the government into this confrontation. I figure it's a simpler "Oh look, we now have a good opportunity to make that announcement that we were worried about." Considering it's probably the same high-level decision makers on both choices it doesn't need a board meeting. And yes they're absolutely free to do what they want, but they're also not blind to how the public will view their decisions.

My theory is that Anthropic has been wanting to make this change and doing it now while they’re making a (leaked to the) public stand in the name of ethics was a good opportunity.

Seems like the two main threats are Defense Production Act and Supply Chain Risk. I'd assume Anthropic would sue if either were invoked. I could imagine Supply Chain Risk being easier to push back on because it's pretty clearly being used punitively rather than because of an actual risk. DPA might be a bit harder to push back on if the banned functionality (i.e. mass surveillance and autonomous weapons) exists in the LLM itself and it's just a matter of disabling external checks. If the banned functionality is baked into the training data/weights directly they could probably push back on the DPA by saying the functionality isn't something they can reasonably create.

Only other precedent I can think of in the case where pushback fails is Lavabit with Edward Snowden's email, but I feel like Anthropic is too big to "fail" in the same way Lavabit did to avoid complying. The penalty for refusing to comply with the Defense Production act is $10k and/or a year in prison, but I think if the government actually pursued that they would burn a bunch of bridges and Amodei would be a folk hero.

Yeah this standoff is worth at least 10 Super Bowl ads in good publicity. The Pentagon is saying "Claude is the best so we need to use it but you need to stop acting ethically". I'm almost wondering if someone in the administration has a stake in Anthropic because this is such a boost.

Their threat to label it a supply chain risk also feels toothless because they've basically admitted that using Claude is a benefit, so by their own logic they're be shooting themselves in the foot to ban contractors from using it.

What I was getting at is that the current API is still a generic inference endpoint, but with OAuth instead of an API key. What I'm suggesting is that they move some of the client logic up to the OAuth endpoint so it's no longer a generic inference endpoint (e.g. system prompt is static, context management is done on the server, etc). I assume they can get it to a point that it's no longer useful for a general purpose client like OpenClaw

I feel like a lot of this would go away if they made a different API for the “only for use with our client” subscriptions. A different API from the generic one, that moved some of their client behaviors up to the server seems like it would solve all this. People would still reverse engineer to use it in other tools but it would be less useful (due to the forced scaffolding instead of entirely generic completions API) and also ease the burden on their inference compute.

I’m sure they went with reusing the generic completions API to iterate faster and make it easier to support both subscription and pay-per-token users in the same client, but it feels like they’re burning trust/goodwill when a technical solution could at least be attempted.

I’ve thought an interesting outcome might be that it’s not even that there’s a binary generated. It’s just user input -> machine code LLM -> CPU. Like the only binary would be the LLM itself and it’s essentially mimicking software live. The paper “Diffusion as a Model of Environment Dream” (DIAMOND) is close to what I’m thinking, where they have a diffusion model generate frames of a game, updating with user input, but there’s no actual “game” code it’s just the model.

https://diamond-wm.github.io/

Like you’d have a machine code LLM that behaves like software but instead of a static binary being executed it’s just the LLM itself “executing” on inputs and precious state. I’m horrible at communicating this idea but hopefully the gist is there.

What you’re describing is called “rollback netcode”. It’s a pretty cool chunk of theory, usually used for fighting games which are extremely sensitive to latency. This explainer has some nice graphic demos

https://bymuno.com/post/rollback