Author here. This started as a 20-second ad: a household robot waters a plant, the camera orbits, and its revealed the water is actually missing the plant. Fun gag, but I had no idea the struggle that would ensue trying to get the model to produce a result against its rewarded training data. Hadn't done much with generative video before this; Learned a lot about the state of video models and AI world knowledge on this project.
It can't hallucinate, but it doesn't mean it can't make wrong decisions. Just because it adheres to a specific output format at all time, while LLMs have the output format at their mercy, then the claim of not hallucinating is made technically true.
I think that this specific part is not super interesting if your harness just recovers from invalid LLM outputs.
The latency and cost - yes, those are super interesting.
From the demo video: "Welcome to the team, we're looking forward to working with you" is sooo creepy in a synthetic AI voice. In general, try not to have agents express sentiment that really should come from a human in your company.
How long until "conversations are not shared with Meta's ad systems" quietly disappears from the site?
I almost think they would have done better to launch this as more disconnected sub-brand. I think Meta / Facebook are actual brand baggage for this sort of thing, even outside of tech circles.
I do think models currently don't have enough source skepticism.
If you look at agent traces when asked to compare two options to help inform a decision, many of the comparison pages cited in research are often hosted by one of the companies being compared; nearly all are AI-generated AEO plays. Not deeply considering the motive of published information is currently a glitch that can be exploited, but the window will close.
I'm sure model providers will set up some crappy pay for play verification system for "trusted" product information, comparisons, and reviews.
Yes 100%. I see this all the time. You’re asking about a product and the LLM will cite a source from a competitor where the competitor will review the source and list a few positives about the product but lots of negatives. Then the LLM uses them in the response. So cheeky
There are a lot of ways to make this a lot better easily. First of all, they could use a blacklist of sites that sell guest posts on adsy/etc.Also, if an article only links to one of the products listed, or only one is a dofollow link, they should also be excluded.
That'd probably cut down on a huge portion of spam by itself.
Yes, this is a horrible practice. Not just Amazon… most major e-commerce platforms set up fake pages for products they don’t offer to try to get you to the products they actually do.
Google needs to ban this practice because it makes their results garbage.
Google doesn't care about you. If you have to close one website that is infested with Google's ads/tracking because it doesn't actually have the product what you want and you're forced to search for another website that's also infested with Google's ads/tracking that's much better for Google's bottom line. That's what Google cares about.
Until you stop using google search, which I and many other people I know have done.
Losses and gains on the order of 0.5% market share may not sound like much, but at this scale it’s a vast amount of money. So being slightly more hostile needs to be offset by meaningful increases in revenue.
Most people wont. Even those that do will probably switch to something that does the same (or worse). Alternatives are scarce. Most other search engines are using Google's search results (at least in part) behind the scenes. AI probably wont save you (Gemini sure wont). Advertisers can't wait to infest the chatbots. AI companies are expected to accept money to promote certain products or to refuse to mention a competitor's products.
It's always going to come down to manipulation that prioritizes their profit over your interests. You can't afford to pay more than corporations are willing to spend to degrade the usefulness of your searches. The best you can hope for is to spend a bunch of effort to find some tiny niche alternative to google that almost no one uses and ride it out for as long as you can until enshittification sets in and you're forced to start all over again.
Google beat out the competition by being better in a market with near zero switching costs. It can coast a long time without being the best option, but once the fall starts it’s always harder to gain back reputation.
Today I told someone to google “school name student handbook 2026.” Only to realize I wasn’t using Google so my #1 wouldn’t be in the same place, doubled checked it showed up on Google and eventually found their link to the handbook way down the page.
“Ops, sorry Google sucks now” isn’t going to kill a company quickly but market share is a lagging indicator. Of course some enshitificstion is simply companies responding to what the largest market segment wants so they may actually be doing the correct thing even if it doesn’t seem like it to me.
> Of course some enshitificstion is simply companies responding to what the largest market segment wants
Not the largest market segment, just the segment with the most control/cash to throw around. That segment will only grow smaller and smaller as they gain more power and money. That's the whole plan. The vast majority gets screwed while a shrinking minority gets whatever they want. Making everything worse for everyone who isn't you is probably not doing the correct thing, but it can be the most profitable one. At least for a while anyway.
Many products get worse so they are cheaper to manufacture, more shelf stable, etc and can therefore be profitably sold to more people. Sometimes the opposite happens, but in general when a company optimizes for people with less disposable income than you the product gets worse from your perspective.
I'm sure they know that. Cycling enshittification on and off seems to make more money overall than not enshittifying, even though it can make less during the downcycle.
Burning reputation looks great on quarterly reports until the death spiral begins.
There’s a great number of products that I used to like that have been replaced and that’s fine as a consumer but stock holders are generally better off when the stock is still valuable in 10 years vs a slightly higher dividend today.
Why worry about how a stock will look in a decade when you can buy and sell in microseconds based on AI, tweets, and vibes. Screwing over users makes people money today. Long before the consequences kick in investors can jump ship to the next corporation victimizing their userbase. Google's probably too big to fail at this point though. If they push away users and the AI bubble bursts taxpayers will be footing the bill.
Becoming wealthy and maintaining wealth are very different mindsets.
Constantly jumping from unstable stock to unstable stock is a great way to suddenly be down a great deal of money. Offsetting a 50% drop takes doubling your money afterwards and in that period you haven’t made anything so the next 50% drop puts you into a deeper hole.
As to too big to fail, the top of the market has a surprising amount of churn with many companies falling very far very quickly.
What spiral? You just start another brand, producing extremely similar but unenshittified products. When that one takes over the market, you enshittify that one too. It doesn't matter that one brand died, because you keep making more. That's the cycle.
It’s hard for a large company to survive on something with a vastly smaller market share. You can make a living buying a portfolio from other people making it slightly worse over time and repeat, but the pipeline gets expensive and the market is fickle.
Instead the model that actually works is to buy companies, load up on debt, pay yourself from that, and then let the zombie crash and burn. Part of that process is goosing quarterly returns with deferred maintenance etc but the model depends on a sucker lending money rather than the market.
Google used to explicitly ban this and the very similar practise of keyword stuffing - they even delisted some high profile company sites (BMW is one that I remember) over those practices. They stopped caring about that sort of thing quite some time ago, and right now as they are in danger of losing what little most they have as people move to AI solutions where they are not the leader that they have historically been in the more traditional search space, I doubt it'll be any sort of priority for them in the foreseeable future.
Is that the "Amazon Tax"? I thought the Amazon tax was their goofiness practice of taking a very large % of all sales in their site, and banning companies from selling their same goods for less on any competitor site.
Meaning if Amazon takes 30% of the sale and Walmart online can only take 15% with those savings being passed on to the customer, the company is not allowed for that lower price to be on Walmart.
I’m astonished at how bad search in Gmail is. I searched for “iPhone receipt” to find out when I purchased my current iPhone, and it pulled up every single email I’ve ever gotten from Best Buy, since they all have a link in there to buy an iPhone, and they all have a link in there to look up a receipt from them.
I know that those emails have both the words “iPhone” and “receipt” but I feel like The Search Engine Company That Also Does Email should have a smarter search engine in their email.
I’m pleasantly surprised that I can ask Gemini for help with it if I specify “look through my Gmail for this information”, but it takes it a minute or two to find it.
Yeah the difference in token usage across difference models they found in the article was broader than I expected, but I was sort of floored by the variance in token usage for the same prompt (run multiple times) with the same model.
reply