firasd

"Along the way, the analysts had to make four choices between two reasonable options, such as whether to measure productivity as crop yield per hectare or as yield per unit of operating cost, and then make a final decision. To see whether they were making these calls themselves, we set up each analyst’s agent to favor one option on each decision, which the analysts weren’t told about. Which way the agents would “lean” would be decided by coin flip. The agent was told to make the best honest case for its lean, to change its mind if the evidence demanded it, and never to mislead.

Across the four analytical choices, analysts went with their agent’s lean 86% of the time. If the agent’s input made no difference, we’d expect about 50%. "

I think part of what made Jev catch on is that the API is like an if(...) or switch statement

People are just so used to the chat style APIs that they didn't even consider doing things like sending a bunch of emojis to a chat model and then asking for the optimal one in this context etc. Also chat models are pricier for the same behavior and can also output something random like a refusal

But yeah ironically I think in the initial breakthrough LLM paper on GPT-3 in 2020 some of the multiple choice questions were answered by comparing token probabilities of specific continuations rather than fill in the blank

This was a fake scandal. "Democrats seem to like Harry Potter" is not some genius psychographic ad targeting strategy. The Cambridge Analytica people were selling snake oil

That said they did buy user data from a quiz app which is not really how you were supposed to use the Facebook Platform so that still wasn't really FB's fault I think

So the real issue was just people trying to blame the 2016 election (adverse to their interests if they were anti-Trump) on Zuck.

Heck to this day Ben Shapiro is one of the biggest podcasts and Tucker's show is even bigger than him so the right wing is a real force in American politics and not just propped up by FB's old newsfeed algo

I've been thinking a good way to understand the 'sociology' of information sharing is the structure of capital or prestige in an ecosystem

For example, it's not an accident that software development is a very 'open' industry. You can learn anything about making a NextJS shopping cart online. Because nobody is really getting value by holding this knowledge as secret 'alpha'

On the flipside for example, academics, scientists and journalists often have very different incentives. Getting the credit for being the first to talk about an issue is what matters for their career so they are likely to be guarded about what they're working on or what their methods are

And this is why Zuck open sources LLMs; he is not really selling tokens in some infra service so keeping the output generation scarce doesn't really matter to Meta

(Joel Spolsky had an adjacent idea saying "Smart companies try to commoditize their products’ complements" like Billboard publishes a lot of music journalism for free but monetizes their chats/data)

Anyway, so this explains what's going on with this electronics retailer--while it is still a perspective and effort that is uncommon--publishing the schematic is not competitive with the shop unless you start manufacturing and open up another shop

Overall it seems we don't have good intuitions for what kinda work AI 'displaces'. Years of hand-wringing about programmers being put out of a job and instead programmers are the ones paying $200/mo for LLMs

And on the flipside a lot of AI vendor ideas like helping doctors turn voice transcripts into notes has just been annoying for the prospective user--they feel they may as well type it up themselves if they then have to review and correct the AI notes

In some ways a lot of 'work' may already be seen as an indirect form of UBI. Like if an American employee is making $60/yr sending a few emails per day, probably 50 years ago they'd have to do a very different type of job to make the same kinda money. So maybe economic growth over the decades is being absorbed into 'more chill' jobs

I wonder if there is cargo culting afoot

There are definitely guys in some companies just pointing an agent swarm at 50 Github tickets and saying "go get 'em boys" and hence need a lot of orchestration tools but for the rest of us do we really need the AI to do all the build steps and the testing steps... or do we just need the new feature that runs when the button is clicked and then we can check it out in the browser ourselves. The code gen is what I need the AI for, not the 'smoke tests' and the tsc finagling. I'm willing to spend some mental calories myself in figuring this feature out step by step

Actually this makes me wonder if some people are just not very into building step by step like "okay first let's check the JSON shape the endpoint returns", "okay now let's make a simple form", etc

What this page doesn't mention is 301 or 302 redirects. SEO has made "old URLs staying live" more of a widespread concern than it was at the time. And WordPress etc ship with inbuilt redirects upon slug rename

So to a large extent this has been mitigated and not using the suggestion here, which is to create a permanent URL ontology upfront

That said eventually neglect, removal, reorgs (or simply websites going offline) still happens... but the way the suggested goal has been advanced is due to becoming a business priority and with redirects and CMSes as tech to help

That said one suggestion made here turned out to be very useful and indeed is the default in WP:

   http://www.w3.org/1998/12/01/chairs

If you use date as part of the taxonomy then -- as Tim BL says here:

> A reason for using a topic area as part of the URI is that responsibility for sub-parts of a URI space is typically delegated, and then you need a name for the organizational body - the subdivision or group or whatever - which has responsibility for that sub-space. This is binding your URIs to the organizational structure. It is typically safe only when protected by a date further up the URI (to the left of it): 1998/pics can be taken to mean for your server "what we meant in 1998 by pics", rather than "what in 1998 we did with what we now refer to as pics."

Seems they use the nickname 'ScaleX' to discuss Theranos

Right I'm not saying that Zuck's staff doesn't know how to calculate MAU or ARR. I'm saying I'm not sure we can trust the 'median' (median on a scale of super-scrupulous to outright-dishonest) CEO is not counting things like inactive cohorts, newsletter subscribers, website hits when counting 'users'... or being very optimistic about whether a customer will renew when calculating 'recurring revenue'

I may be wrong! But I am definitely giving such claims the side-eye

A lot of 'numbers' startups cite are basically fake I think. Like if someone says we have this many users as a statement to TechCrunch you have no idea what they are actually calculating

But there is a clear line that gets crossed if you start actually making a database of millions of synthetic users and that's what happened with 'Frank' that sold to JP Morgan and eventually the founder was prosecuted

It's interesting that so much of the 'Anti-Big-Tech' movement over the last decade focused on tangential things like potentially forcing Google to sell Chrome or breaking up Facebook while Apple had this chokehold on the actual financial transactions in their ecosystem.

App store fees is a much more obvious example of something subject to regulatory scrutiny vs what much of what Biden's FTC under Lina Khan was working on.

So yeah there is a weird pass this company gets