Hacker News

Agreed. Gemini 3 Pro for me has always felt like it has had a pretraining alpha if you will. And many data points continue to support that. Even as flash, which was post trained with different techniques than pro is good or equivalent at tasks which require post training, occasionally even beating pro. (eg: in apex bench from mercor, which is basically a tool calling test - simplifying - flash beats pro). The score on arc agi2 is another datapoint in the same direction. Deepthink is sort of parallel test time compute with some level of distilling and refinement from certain trajectories (guessing based on my usage and understanding) same as gpt-5.2-pro and can extract more because of pretraining datasets.

(i am sort of basing this on papers like limits of rlvr, and pass@k and pass@1 differences in rl posttraining of models, and this score just shows how "skilled" the base model was or how strong the priors were. i apologize if this is not super clear, happy to expand on what i am thinking)

Per BalatroBench, gemini-3-pro-preview makes it to round (not ante) 19.3 ± 6.8 on the lowest difficulty on the deck aimed at new players. Round 24 is ante 8's final round. Per BalatroBench, this includes giving the LLM a strategy guide, which first-time players do not have. Gemini isn't even emitting legal moves 100% of the time.

It's trained on YouTube data. It's going to get roffle and drspectred at the very least.

Google has a library of millions of scanned books from their Google Books project that started in 2004. I think we have reason to believe that there are more than a few books about effectively playing different traditional card games in there, and that an LLM trained with that dataset could generalize to understand how to play Balatro from a text description.

Nonetheless I still think it's impressive that we have LLMs that can just do this now.

DeepSeek hasn't been SotA in at least 12 calendar months, which might as well be a decade in LLM years

How does it do on gold stake?

But... there's Deepseek v3.2 in your link (rank 7)

> . I don't think there are many people who posted their Balatro playthroughs in text form online

There are *tons* of balatro content on YouTube though, and it makes absolutely zero doubt that Google is using YouTube content to train their model.

> Most (probably >99.9%) players can't do that at the first attempt

Eh, both myself and my partner did this. To be fair, we weren’t going in completely blind, and my partner hit a Legendary joker, but I think you might be slightly overstating the difficulty. I’m still impressed that Gemini did it.

François Chollet, creator of ARC-AGI, has consistently said that solving the benchmark does not mean we have AGI. It has always been meant as a stepping stone to encourage progress in the correct direction rather than as an indicator of reaching the destination. That's why he is working on ARC-AGI-3 (to be released in a few weeks) and ARC-AGI-4.

His definition of reaching AGI, as I understand it, is when it becomes impossible to construct the next version of ARC-AGI because we can no longer find tasks that are feasible for normal humans but unsolved by AI.

Yes, but benchmarks like this are often flawed because leading model labs frequently participate in 'benchmarkmaxxing' - ie improvements on ARC-AGI2 don't necessarily indicate similar improvements in other areas (though it does seem like this is a step function increase in intelligence for the Gemini line of models)

Here's a good thread over 1+ month, as each model comes out

https://bsky.app/profile/pekka.bsky.social/post/3meokmizvt22...

tl;dr - Pekka says Arc-AGI-2 is now toast as a benchmark

Agreed. I love the elegance of ARC, but it always felt like a gotcha to give spatial reasoning challenges to token generators- and the fact that the token generators are somehow beating it anyway really says something.

The average ARC AGI 2 score for a single human is around 60%.

"100% of tasks have been solved by at least 2 humans (many by more) in under 2 attempts. The average test-taker score was 60%."

https://arcprize.org/arc-agi/2/

Wouldn't you deal with spatial reasoning by giving it access to a tool that structures the space in a way it can understand or just is a sub-model that can do spatial reasoning? These "general" models would serve as the frontal cortex while other models do specialized work. What is missing?

Why 5-10 years?

At current rates, price per equivalent output is dropping at 99.9% over 5 years.

That's basically $0.01 in 5 years.

Does it really need to be that cheap to be worth it?

Keep in mind, $0.01 in 5 years is worth less than $0.01 today.

A grad student hour is probably more expensive…

What’s reasonable? It’s less than minimum hourly wage in some countries.

That's not a long time in the grand scheme of things.

The puzzles are calibrated for human solve rates, but otherwise I agree.

I for one welcome our new AI overlords.

Singularity or just Chinese New Year?

Please use the term “Lunar New Year” instead of “Chinese New Year,” as the lunar calendar is a respected tradition in many Asian countries. For example, both California and New York use the term “Lunar New Year” in their legislation.

> Also, I don't understand the comments about Google being behind in agentic workflows.

It has to do with how the model is RL'd. It's not that Gemini can't be used with various agentic harnesses, like open code or open claw or theoretically even claude code. It's just that the model is trained less effectively to work with those harnesses, so it produces worse results.

There are hints this is a preview to Gemini 3.1.

Most ridiculous valuation in the history of markets. Cant wait to watch these compsnies crash snd burn when people give up on the slot machine.

I must be holding these things wrong because I'm not seeing any of these God like superpowers everyone seem to enjoy.

> I don’t think it’s hyperbolic to say that we may be only a single digit number of years away from the singularity.

We're back to singularity hype, but let's be real: benchmark gains are meaningless in the real world when the primary focus has shifted to gaming the metrics

I agree. On top of that, in true Google style, basic things just don't work.

Any time I upload an attachment, it just fails with something vague like "couldn't process file". Whether that's a simple .MD or .txt with less than 100 lines or a PDF. I tried making a gem today. It just wouldn't let me save it, with some vague error too.

I also tried having it read and write stuff to "my stuff" and Google drive. But it would consistently write but not be able to read from it again. Or would read one file from Google drive and ignore everything else.

Their models are seriously impressive. But as usual Google sucks at making them work well in real products.

How can the models be impressive if they switch to Chinese mid-sentence? I've observed those bizarre bugs too. Even GPT-3 didn't have those. Maybe GPT-2 did. It's actually impressive that they managed to botch it so badly.

Google is great at some things, but this isn't it.

It's so capable at some things, and others are garbage. I uploaded a photo of some words for a spelling bee and asked it to quiz my kid on the words. The first word it asked, wasn't on the list. After multiple attempts to get it to start asking only the words in the uploaded pic, it did, and then would get the spellings wrong in the Q&A. I gave up.

Agreed on the product. I can't make Gemini read my emails on GMail. One day it says it doesn't have access, the other day it says Query unsuccessful. Claude Desktop has no problem reaching to GMail, on the other hand :)

Sadly true.

It is also one of the worst models to have a sort of ongoing conversation with.

Their models are absolutely not impressive.

Not a single person is using it for coding (outside of Google itself).

Maybe some people on a very generous free plan.

Their model is a fine mid 2025 model, backed by enormous compute resources and an army of GDM engineers to help the “researchers” keep the model on task as it traverses the “tree of thoughts”.

But that isn’t “the model” that’s an old model backed by massive money.

OpenAI is the best thing that happened to Google apparently.

Wartime Google gave us Google+. Wartime Google is still bumbling, and despite OpenAI's numerous missteps, I don't think it has to worry about Google hurting its business yet.

> I feel like most of human endeavor will soon be just about trying to continuously show that AI's don't have AGI.

I think you overestimate how much your average person-on-the-street cares about LLM benchmarks. They already treat ChatGPT or whichever as generally intelligent (including to their own detriment), are frustrated about their social media feeds filling up with slop and, maybe, if they're white-collar, worry about their jobs disappearing due to AI. Apart from a tiny minority in some specific field, people already know themselves to be less intelligent along any measurable axis than someone somewhere.

Soon they can drop the bioweapon to welcome our replacement.

"AGI" doesn't mean anything concrete, so it's all a bunch of non-sequiturs. Your goalposts don't exist.

Anyone with any sense is interested in how well these tools work and how they can be harnessed, not some imaginary milestone that is not defined and cannot be measured.

I find Gemini's web page much snappier to use than ChatGPT - I've largely swapped to it for most things except more agentic tasks.

Google privacy cred is ... excellent? The worst data breach I know of them having was a flaw that allowed access to names and emails of 500k users.

> Gemini's UX ... is the worst of all the AI apps

Been using Gemini + OpenCode for the past couple weeks.

Suddenly, I get a "you need a Gemini Access Code license" error but when you go to the project page there is no mention of this or how to get the license.

You really feel the "We're the phone company and we don't care. Why? Because we don't have to." [0] when you use these Google products.

PS for those that don't get the reference: US phone companies in the 1970s had a monopoly on local and long distance phone service. Similar to Google for search/ads (really a "near" monopoly but close enough).

0 - https://vimeo.com/355556831

You mean AI Studio or something like that, right? Because I can't see a problem with Google's standard chat interface. All other AI offerings are confusing both regarding their intended use and their UX, though, I have to concur with that.

Gemini is completely unusable in VS Code. It's rated 2/5 stars, pathetic: https://marketplace.visualstudio.com/items?itemName=Google.g...

Requests regularly time out, the whole window freezes, it gets stuck in schizophrenic loops, edits cannot be reverted and more.

It doesn't even come close to Claude or ChatGPT.

You sound like Russ Hanneman from SV

I looked at the file name but not the document title (specifically because I was wondering if this is 3.1). Good spot.

edit: they just removed the reference to "3.1" from the pdf

That's odd considering 3.0 is still labeled a "preview" release.

The rumor was that 3.1 was today's drop

Part of OpenRouter's appeal to me is precisely that it is a middle man. I don't want to create accounts on every provider, and juggle all the API keys myself. I suppose this increases my exposure, but I trust all these providers and proxies the same (i.e. not at all), so I'm careful about the data I give them to begin with.

i think this is the right answer

edit: i don't know how this is meaningfully different from 3

its interesting that opus 4.6 added a paramter to make it think extra hard.

Well, one thing i know for sure: it reliably misplaces parentheses in lisps.

> Trouble is some benchmarks only measure horse power.

IMO it's the other way around. Benchmarks only measure applied horse power on a set plane, with no friction and your elephant is a point sphere. Goog's models have always punched over what benchmarks said, in real world use @ high context. They don't focus on "agentic this" or "specialised that", but the raw models, with good guidance are workhorses. I don't know any other models where you can throw lots of docs at it and get proper context following and data extraction from wherever it's at to where you'd need it.

It's a giant game of leapfrog, shift or stretch time out a bit and they all look equivalent

The idea is that each subagent is focused on a specific part of the problem and can use its entire context window for a more focused subtask than the overall one. So ideally the results arent conflicting, they are complimentary. And you just have a system that merges them.. likely another agent.

You just pipe it to another agent to do the reduce step (i.e. fan-in) of the mapreduce (fan-out)

It's agents all the way down.

Start with 1024 and use half the number of agents each turn to distill the final result.

Not very likely?

ARC-AGI-3 has a nasty combo of spatial reasoning + explore/exploit. It's basically adversarial vs current AIs.

The AGI bar has to be set even higher, yet again.

wow solving useless puzzles, such a useful metric!

Its possibly label noise. But you can't tell from a single number.

You would need to check to see if everyone is having mistakes on the same 20% or different 20%. If its the same 20% either those questions are really hard, or they are keyed incorrectly, or they aren't stated with enough context to actually solve the problem.

It happens. Old MMLU non pro had a lot of wrong answers. Simple things like MNIST have digits labeled incorrect or drawn so badly its not even a digit anymore.

Everyone is already at 80% for that one. Crazy that we were just at 50% with GPT-4o not that long ago.

It's correlated with the ability to solve logic puzzles.

It's also interesting because it's very very hard for base LLMs, even if you try to "cheat" by training on millions of ARC-like problems. Reasoning LLMs show genuine improvement on this type of problem.

how would we actually objectively measure a model to see if it is AGI if not with benchmarks like arc-AGI?

If anyone trains a model on https://simonwillison.net/tags/pelican-riding-a-bicycle/ they're going to get some VERY weird looking pelicans.

For every combination of animal and vehicle? Very unlikely.

The beauty of this benchmark is that it takes all of two seconds to come up with your own unique one. A seahorse on a unicycle. A platypus flying a glider. A man’o’war piloting a Portuguese man of war. Whatever you want.

You can always ask for a tyrannosaurus driving a tank.

I've heard it posited that the reason the frontier companies are frontier is because they have custom data and evals. This is what I would do too

It's not actually, look up some photos of the sun setting over the ocean. Here's an example:

https://stockcake.com/i/sunset-over-ocean_1317824_81961

It depends, if you meant from a human coding an SVG "manually" the same way, I'd still say this is excellent (minus the reflection issue). If you meant a human using a proper vector editor, then yeah.

Indeed. And when you factor in the amount invested... yeah it looks less impressive. The question is how much more money needs to be invested to get this thing closer to reality? And not just in this instance. But for any instance e.g. a seahorse on a bike.

I'll agree to disagree. In any thread about a new model, I personally expect the pelican comment to be out there. It's informative, ritualistic and frankly fun. Your comment however, is a little harsh. Why mad?

Eh, i find it more of a not very informative but lighthearted commentary

It being a terrible benchmark is the bit.

I disagree. The task asks for an SVG; which is a vector format associated with line drawings, clipart and cartoons. I think it's good that models are picking up on that context.

In contrast, the only "realistic" SVGs I've seen are created using tools like potrace, and look terrible.

I also think the prompt itself, of a pelican on bicycle, is unrealistic and cartoonish; so making a cartoon is a good way to solve the task.

The request is for an SVG, generally _not_ the format for photorealistic images. If you want to start your own benchmark, feel free to ask for a photorealistic JPEG or PNG of a pelican riding a bicycle. Could be interesting to compare and contrast, honestly.

At the very least gemini 3's flyer claims 1T parameters.

> Its really weird how you all are begging to be replaced by llms, you think if agentic workflows get good enough you're going to keep your job? Or not have your salary reduced by 50%?

The computer industry (including SW) has been in the business of replacing jobs for decades - since the 70's. It's only fitting that SW engineers finally become the target.

I agree with you and have similar thoughts (maybe, unfortunately for me). I personally know people who outsource not just their work, but also their life to LLMs, and reading their exciting comments makes me feel a mix of cringe, fomo and dread. But what is the engame for me and you likes, when we finally would be evicted from our own craft? Stash money while we still can, watching 'world crash and burn', and then go and try to ascend in some other, not yet automated craft?

I think a lot of people assume they will become highly paid Agent orchestrators or some such. I don't think anyone really knows where things are heading.

Most folks don't seem to think that far down the line, or they haven't caught on to the reality that the people who actually make decisions will make the obvious kind of decisions (ex: fire the humans, cut the pay, etc) that they already make.

You don't hate AI, you hate capitalism. All the problems you have listed are not AI issues, its this crappy system where efficiency gains always end up with the capital owners.

Or we just end capitalism.

French revolution style.

shrugs

It would make more sense to me if it had never been awesome.

@dang will often replace the post url & merge comments

HN guidelines prefer the original source over social posts linking to it.

What do you use it for and why? Genuinely curious