HNHacker News
TopNewBestAskShowJobs

jjani

1,330 karma · joined March 1, 2025

submissionscomments
jjani··on Meta suppressed research on child safety, employees say
Facebook still has a pedo problem because they still haven't hired 1 million employees to deal with it. Amazon would still have a logistics problem if they hadn't hired 1 million employees to move boxes around.
jjani··on Ex-WhatsApp cybersecurity head says Meta endangered billions of users
Central Asia is Telegram as well.
jjani··on Incident Report for Anthropic
We talked about this a few weeks ago so it can't be the first time you're hearing about this :) [1] You hadn't heard about it before because it really only affected API customers running services including calls that required output of 2.5k+ (rough estimate) tokens in a single message. Which is pretty much just the small subset of AI founders/developers that are in the long-form content space. And then out of those, the ones using Sonnet 3.5 at the time for these tasks, which is an even smaller number. Back then it wasn't as favoured yet as it is now, especially for content. It's also expensive for high-output tasks, so only relevant to high-margin services, mostly B2B. Most of us in that small group aren't posting online about it - we urgently work to work around it, as we did back then. Still, as I showed you, some others did post about it as well.

The only explanations were either internal system prompt changes, or updating the actual model. Since the only sharply different evals were those expecting 2.5k+ token outputs with all short ones remaining the same, and the consistency of the change was effectively 100%, it's unlikely to have been a stealth model update, though not impossible.

[1] https://news.ycombinator.com/item?id=44844311

jjani··on ICEBlock handled my vulnerability report in the worst possible way
Is this guy fishing for a job at DOGE? Otherwise I'm not sure what could explain why he's acting in pretty much the "worst practice" manner possible when doing security reports. Even stuff like the literal teens doing the Burger King (? iirc) and Monster energy reports that got posted here recently, while flawed, were still way better than this.
jjani··on Incident Report for Anthropic
They already have a track record of messing with internal system prompts (including those that affect the API) which obviously directly change outputs given the same prompts. So in effect, they've already been messing with the models for a long time. It's well known among founders who run services based on their products that this happened, everyone who does long output saw the same. It happened around November last year. If you had a set of evals running that expected an output of e.g. 6k tokens in length on 3.5 Sonnet, overnight it suddenly started cutting off at <2k, ending the message with something like "(Would you like me to continue?)". This is on raw API calls.

Never seen or heard of (from people running services at scale, not just rumours) this kind of API behaviour change for a the same model from OpenAI and Google. Gemini 2.5 Pro did materially change at time of prod release despite them claiming they had simply "promoted the final preview endpoint to GA", but in that case you can give them the benefit of it being technically a new endpoint. Still lying, but less severe.

jjani··on Meta suppressed research on child safety, employees say
> It's just also facially incorrect - Amazon does hire warehouse workers. Facebook does not hire enough content monitors.

This is the whole point. Amazon had to hire hundreds of thousands of warehouse workers to scale. They have 1.5 million employees. Facebook is capable of doing the same. The idea that they "can't scale" if they have to stop unloading their negative externalities is absurd. Amazon scaled, while hiring 1.5 million employees. Meta can scale and do the same.

jjani··on Google's new AI mode is good, actually
You've been asking things that are verbatim in the first few links. Ask anything that isn't and it will confidently hallucinate.
jjani··on Google's new AI mode is good, actually
IME Gemini has a lot more non-tech mindshare than Claude. For tech it's of course the opposite.
jjani··on Google's new AI mode is good, actually
Gemini models have always been the strongest at image understanding. We're talking like 1.5 years now. It's also better at a good number of non-English/European languages, this too has been ever since 1.5 pro.
jjani··on Meta suppressed research on child safety, employees say
Anything without a feed backed by a recommender system.

Front Porch Forum is one example of a relatively good social network. It's made possible by the founders not aiming to become billionaires. This is another necessary property of basically anything good.

jjani··on Meta suppressed research on child safety, employees say
Going to prison would not be enough, he'd be pardoned very quickly and there'd be very little of a discouraging effect on other megacorps with similar ethics.
jjani··on Meta suppressed research on child safety, employees say
I'm flabbergasted whenever I read this argument.

It's like saying Amazon's business is not scalable because they need warehouse workers.

jjani··on Garmin beats Apple to market with satellite-connected smartwatch
IME it very much doesn't. Bluetooth headphones/earphones in the airplane while on airplane mode are now part of near every flight.
jjani··on Garmin beats Apple to market with satellite-connected smartwatch
"As accurate as Oura" or "as is Oura"? The missing word is adding some ambiguity here :)
jjani··on Purikura: The Japanese Grandmother of the Selfie
I think this boom started here in Korea, I remember they were already growing quickly here back in 2018 (7 years ago, scary how time flies). They're now exporting them all over the world, I talked to someone who helps run one of the chains and there's a lot of demand from South-East Asia.
jjani··on I kissed comment culture goodbye
I didn't even remember whether it had a strikethrough or not but recognized it. It's nothing to do with bots, it's simple human pattern recognition. That rough sequence of characters is really only used in comments for that meme. Plus that it would've been placed in a certain context, making it even easier to recognize, which your comment didn't even have but it was still recognizable.
jjani··on European Commission fines Google €2.95B over abusive ad tech practices
Being made to follow laws is being pushed around?

If Trump makes Google not pay the fine you think that will have no negative side-effects? His actions have been incredibly positive for European tech companies, 5 years ago the only ones that even considered not going for the US options were a few companies in Germany. Demand has skyrocketed and making Google not pay this would give it a huge boost.

jjani··on European Commission fines Google €2.95B over abusive ad tech practices
I don't think I expressed anything about their likelihood of payment - just that they won't stop doing business in the EU.

I think your scenario is a real possibility, but ironically one that would cost the US a lot more than it gains. It's really playing with fire, running the risk of even just 1% of EU businesses and consumers opting for EU services over US ones. And just that 1% represents far more than all of the yearly fines to Google/Meta/Apple combined.

jjani··on European Commission fines Google €2.95B over abusive ad tech practices
Well written. The comparison with physical goods as you're making it is one I'm a big fan of, and should be made much more often.

It's laughable that tarriffs and import taxes only apply to physical goods. If the EU had even an ounce of self-respect, the second the US came out with the tarriffs, they would've come out and said:

"We think this is a fantastic idea by Mr. Trump. Aligned with his views, we are instituting accompanying digital tarriffs to fix the digital trade defecit. We're sure he'll agree that the trade balance should be corrected in both the physical as well as digital worlds".

And that's why the US is so mad at the likes of Brazil - finally, after decades of getting rinsed, countries are starting to take (wholly insufficient) measures here and rightly instituting the equivalent of digital tarriffs.

jjani··on European Commission fines Google €2.95B over abusive ad tech practices
The EU is still a massive profit center for these companies. Over 2025 Alphabet's revenue was around $170B in US and $100B in EMEA. Imagine if Google couldn't operate in half of the US, and how impactful that would be. Yet EMEA revenue is higher than that.
jjani··on European Commission fines Google €2.95B over abusive ad tech practices
China is a special case. As the government has effectively total control, even having 30% market share as Google doesn't mean it's a profitable venture, or that the government hasn't said that it will let it remain one - and there's really no way that it would've given that it's relatively easily replaceable critical infrastructure. Search engines aren't as hard as Googlers like to believe, it was substituted by Baidu succesfully. Think about it - which B2C non-Chinese software (not HW) services are still big in China? Anything besides OSes?
jjani··on European Commission fines Google €2.95B over abusive ad tech practices
My yearly earnings are revenue, not profit.
jjani··on GLM 4.5 with Claude Code
Would make very little business sense at this point - currently they have an effective monopoly on routing. Hosting would just make them one provider among a few dozen. It would make the other providers less likely to offer their services through openrouter. It would come with lots of concerns that openrouter would favor routing towards their own offerings. It would be a huge distraction to their core business which is still rapidly growing. Would need massive capital investment. And another thousand reasons I haven't thought of.
jjani··on European Commission fines Google €2.95B over abusive ad tech practices
It's 15% of their yearly net profit in the region. Not even revenue.

3bn sounds like a lot because we haven't gotten used to the absurd profit levels that these monstrosities have reached.

jjani··on European Commission fines Google €2.95B over abusive ad tech practices
US tech does everything it can to forcefully monopolize a market; US tech gets fined by countries who care about it.
jjani··on European Commission fines Google €2.95B over abusive ad tech practices
Going to pre-empt the comments that always pop up in these topics saying "Google/Meta/Apple will just leave the EU at this rate": Google still has around $20 billion yearly reasons to remain active in the EU. Talking Europe yearly net profit here, post-fine. No, they're not going to say "screw this fine, you can take your $20 billion per year, we're leaving!". The second that happens, shareholders will have Sundar's access revoked within the hour.

There is a number of countries where Google has to deal with large levels of protectionist barriers (not the EU, these fines aren't that) and they still operate there. Korea is just one example. Because there's still a lot of money to be made. China isn't a counterexample: Google stopped operating search in China because at that point there was not a lot of money to be made for them in search there.

jjani··on BMW kills Home Assistant integration to “protect security” [video]
A full refund likely not, but compensation likely in countries with proper laws around this like Germany (especially apt since BMW) or Australia.
jjani··on Amazon RTO policy is costing it top tech talent, according to internal document
Vacant ones may drop in value more quickly. Don't want to be seen holding the bag.
jjani··on The staff ate it later
The huge majority of cheese consumed in the US isn't any better than vegan cheese. And yes, the US does have good cheese! It's just a tiny sliver of all cheese consumed.
jjani··on OpenAI says it's scanning users' conversations and reporting content to police
> What's killing us is MBAs and Salesmen.

SamA is not an MBA. He did CS for 2 years and dropped out to build a startup. He's YC personified, and the person most responsible for the phenomenon you're talking about. Take that for what you will.

← PreviousPage 3 of 15Next →