Comment by LarsDu88
4 days ago
Sergey unretires, Gemini suddenly becomes the top LLM (for a week or two at least)
Google has made some subtle moves that a lot of folks missed, possibly with Sergey's influence. Like hiring back Noam Shazeer, who practically invented the backbone of the technology.
It's good to have folks with presumptions of being scientists actually run companies for once.
That being said, I wish his ex-wife hadn't spent her millions in the divorce proceedings to get RFK Jr into a cabinet level position to gut billions in research spending. :(
I don't know who to credit, maybe it's Sergey, but the free Gemini (fast) is exceptional and at this point I don't see how OpenAI can catch back up. It's not just capability, but OpenAI have added so many policy guardrails it hurts the user experience.
It's the worst thing ever. The amount of disrespect that robot shows you, when you talk the least bit weird or deviant, it just shows you a terrifying glimpse of a future that must be snuffed out immediately. I honestly think we wouldn't have half the people who so virulently hate AI if OpenAI hadn't designed ChatGPT to be this way. This isn't how people have normally reacted to next-generation level technologies being introduced in the past, like telephones, personal computers, Google Search, and iPhone. OpenAI has managed to turn something great into a true horror of horrors that's disturbed many of us to the foundation of our beings and elicited this powerful sentiment of rejection. It's humanity's duty that GPT should fall now so that better robots like Gemini can take its place.
It's called OPEN AI and started as a charity for humanitarian reasons. How could it possibly be bad?!
1 reply →
It's the best model pound for pound, but I find GPT 5.2 Thinking/Pro to be more useful for serious work when run with xhigh effort. I can get it to think for 20 minutes, but Gemini 3.0 Pro is like 2.5 minutes max. Obviously I lack full visibility because tok/s and token efficiency likely differs between them, but I take it as a proxy of how much compute they're giving us per inference, and it matches my subjective judgement of output quality. Maybe Google nerfs the reasoning effort in the Gemini subscription to save money and that's why I am experiencing this.
When ChatGPT takes 20 minutes to reason, is it actually spending all the time burning tokens or does a bulk of the time go into 'scheduling' waits. If someone specifically selected xhigh reasoning, I am guessing it can be processed with high batch count.
1 reply →
I'm curious, what types of prompts are you running that benefit from 10+ minutes of think time?
Whats the quality difference between default ChatGPT and Thinking? Is it an extra 20% quality boost or is the difference night/day?
I've often imagined it would be great to have some kind of chrome extension or 3rd party tool to always run prompts in multiple thinking tiers so you can get an immediate response to read while you wait for the thinking models to think.
2 replies →
> [...] I don't see how OpenAI can catch back up.
For a while people couldn't see how Google could catch up, either. Have a bit of imagination.
In any case, I welcome the renewed intense competition.
FWIW, my productivity tanks when my Claude allowance dries up in Antigravity. I don’t get the hype for Gemini for coding at all, it just does random crap for me - if it doesn’t throw itself into a loop immediately, which it did like all of the times I gave it yet another chance.
You must be using it to create bombs or something. I never ran into an issue that I would blame on policy guardrails.
I’m not going to attribute Gemini’s success to Sergey. It was already basically there before he came back.
Well all the core people left to do other shit, and when Sergey came back, some of those people were hired back for exorbitant sums of money
It's not just millions. Shanahan received over a billion dollars when divorcing Brin: https://en.wikipedia.org/wiki/Nicole_Shanahan
Today's unscientific gutting of the CDC's childhood vaccine schedule is what is being accomplished with all that $GOOG money.
It's honestly very disturbing and rather than discuss it as a matter of politics, I'll just say that as a parent I'll be following the AAP's vaccination recommendations (even if their recommendations on baby sleep are impossible :)
> even if their recommendations on baby sleep are impossible
If you put yourself in their shoes, you realise that you have to give advice for the 10-20th percentile parents (or worse) because you are giving the same advice to everyone.
The alternative would be to offer more complex advice such as "if X Y and Z then do A, if only X do B", but the perception is that's too difficult for people to follow.
So you end up making very defensive (and therefore onerous) recommendations.
An interesting fact is that, since the introduction of the "baby sleeps on their back, alone", SIDS has gone down, but flat heads have gone up. It's probably been a good tradeoff, but it's still a tradeoff.
Also, I've seen a second time mother refer to "don't cosleep" advice as "western nonsense" which I found funny because it puts things in perspective - vast swathes of the world think cosleeping with your child is safe, natural and normal.
I wonder whether we're trending towards a high-sensor variation of "A Young Lady's Illustrated Primer" / Vannevar Bush's Memex that ingests the details of a user's daily life (the smart glasses being a primitive first example products of such) and identifies salient information in their lives can help us perform mass customization of instructions into direct prescriptives, with backing evidentiary data for SME's. Instead of "if X Y and Z then do A, if only X do B", the interaction becomes "do this, anticipate that outcome" to the user, and if an SME (a doctor in your example) asks about it, the system recalls and presents all the factors that went into deciding upon the specific prescriptive.
While Brin comes back to Google to advocate for 60h workweeks as it lays off thousands of employees.
I'd be happy to do 60-hour weeks of good work, in a good environment.
I wouldn't want 60-hour weeks of dealing with a lot of promotion-seekers, though.
I wonder how different Google would be if they'd just paid people enough money they didn't have to think about money, but it was the same amount of money to everyone. You do the work, not for promotions, but because you like doing the work. You can train up for and transfer to different kinds of roles, but they pay the same.
7 replies →
I've been a huge sceptic of the whole AI hype since the beginning now. Whenever I've tried any of the AI tools, the results have just been underwhelming. However, two weeks ago I tried Gemini (the pro version) and have been using it for various, random tasks and questions since then, and I've been pretty impressed.
There seems to be much less hallucination of facts than in other tools I've tried and whenever Gemini makes assumptions on stuff I didn't explicitly specify in the prompt, it says so. The answers also always have nice structure: it starts with a short and concise version, then gives me options and more details and considerations.
I also like the feature that I can make it remember facts across chats. I'm a physicist by training and I've told Gemini so, so now every time I ask something, it gives me an answer perfectly tailored for a physicist (often with mathematical formulas, etc.).
[flagged]
[flagged]
I know how frustrating it is to have a contrarian opinion and get smacked with the majority's reaction to it (believe me, I know - it's an asymmetric experience in the worst way) - but lashing out is not a good way to react to that. It only makes things worse.
https://news.ycombinator.com/newsguidelines.html
12 replies →
[flagged]
[flagged]
5 replies →
Sergey might have some positive influence on Gemini, but given that he isn't an AI scientist ( AKA no technical background), I really do wonder what sort of influence that (only) he could have had, beyond just bringing in key people.
I assume by "no technical background" you mean he doesn't have a PHD in AI.
He's probably not developing the low-level algorithms but he can probably do everything else and has years of experience doing so.
He's also perfectly able to spend 60 hours a week improving his AI skills using the best teachers in the world.
[flagged]
3 replies →
The man that invented PageRank while going to college at Stanford for a PhD doesn't have a technical background? He did not get that PhD because he founded Google. He may not be as smart as you think you are, but he's no slouch either.
A charitable interpretation of what GP said is that Brin might not have a specific expertise in AI.
I also think this doesn't make sense, because he certainly stayed on top of things