AlphaEvolve: Gemini-powered coding agent scaling impact across fields

https://deepmind.google/blog/alphaevolve-impact/

by berlianta • 3 months ago

327 148 3 months ago

pingou• 3 months ago

AI improving itself (or at least the architecture it runs on), the singularity is near as they say.

Do we have other examples of AI being used to improve the LLMs, apart for the creation of synthetic data and the testing of the models?

mkw5053• 3 months ago

I feel like the most viral lately is https://github.com/karpathy/autoresearch

NitpickLawyer2• 3 months ago

Do we have other examples of AI being used to improve the LLMs

Yes, last year when they revealed AlphaEvolve they used a previous gemini model to improve kernels that were used in training this gen models, netting them a 1% faster training run. Not much, but still.

lewtun• 3 months ago

Shameless plug: https://huggingface.co/spaces/smolagents/ml-intern

It’s a simple harness around Opus, but with tight integration to Hugging Face infra, so the agent can read papers, test code and launch experiments

westurner• 3 months ago

What are the benchmarks for this, in terms of costs of computation and error; cost to converge?

Re: hyperparameter tuning and autoresearch: https://news.ycombinator.com/item?id=47444581

Parameter-free LLMs would be cool

dinfinity4• 3 months ago

AI improving itself

This is the thing to look for in 2027, imho. All the big AI labs have big projects working on research agents, also specifically into improving AI (duh) and I expect a lot of that to get out of the experimental phases this year.

Next year they actually get to do a lot of work and I think we will see the first big effective architectural change co-invented by AI.

pjmlp• 3 months ago

And then on 2028 we will be selling ice cream at the beach.

cyanydeez4• 3 months ago

The hard part about this is for every few 'WOW', there's a lineage of 'you dumbass'.

I mean, if you can create aharrness to filter these two, sure, singularity away; it's really hard to see how someones gonna do that.

aleksiy1232• 3 months ago

Self improving, doesn’t necessarily imply singularity right?

There still could be hard constraints to make singularity intractable or just such a long time horizon it’s not practical right?

HarHarVeryFunny2• 3 months ago

There is an apples and oranges difference between AI improving itself (becoming more capable) and AI optimizing software that happens to be used for AI training or inference.

A more efficient transformer just costs less to run.

"AI improving AI" would be if one generation of AI designed a next-gen AI that was fundamentally more capable (not just faster/cheaper) than itself. A reptilian brain that could autonomously design a mammalian brain.

Even when hooked up into a smart harness like AlphaEvolve, I don't think LLMs have the creativity to do this, unless the next-gen architecture is hiding in plain sight as an assemblage of parts than an LLM can be coaxed into predicting.

More likely it'll take a few more steps of human innovation, steps towards AGI, before we have an AI capable of autonomous innovation rather than just prompted mashup generation.

Davidzheng• 3 months ago

I don't think there is a fundamental divide between implementation speedups and optimization and algorithmic/architecture optimizations

root_axis• 3 months ago

Singularities are a sign that you have a broken model.

maxothex• 3 months ago

What I'm most curious about is how this translates to messy, real-world codebases without well-defined metrics. Most production software isn't chip design or kernel optimization - it's business logic with unclear success criteria. The infrastructure story is impressive, but I'd love to see how they handle domains where the evaluation function itself is ambiguous.

baq• 3 months ago

RSI is here on the hardware level and on software level. Sprinkle with a couple algorithmic breakthroughs and results are nigh unimaginable.

alecco• 3 months ago

Are Googlers themselves happy using Gemini coding agent instead of Claude Code or Codex? (no snark, I'm really asking)

carbocation• 3 months ago

Last month, Steve Yegge suggested that they are not: https://xcancel.com/Steve_Yegge/status/2043747998740689171

PunchTornado2• 3 months ago

This couldn't be further from the truth

NitpickLawyer2• 3 months ago

He says the problem is that they can't use Claude Code because it's the enemy, and Gemini has never been good enough to capture people's workflows like Claude has, so basically agentic coding just never really took off inside Google. They're all just plodding along, completely oblivious to what's happening out there right now.

This is a bunch of gabagoo. Wrong on so many layers, it's not even worth reading further.

a) goog has agentic coding in both antigravity & cli forms. While it is not at the level of cc + opus, it's still decent.

b) goog has their own versions of models trained on internal code

c) goog has claude in vertex, and most definitely can set it up in secure zones (like they can for their clients) so they'd be able to use claude (at cost) within their own projects.

stormbeard• 3 months ago

Demis Hassabis chimed in on that thread and called it what it is: clickbait.

FrustratedMonky• 3 months ago

There is value in the "eating your own dog food".

If internal staff aren't happy with the tools they build, typically that should drive improvements to their own tools

HardCodedBias2• 3 months ago

He was entirely correct.

He made a follow up after the pushback by GDM.

Google’s businesses are very broad and durable. But Google being the only company in the world without access (except for GDM+labs) to a competent coding agent will take a toll.

We’ll see how long Google can hold out hoping for GDM to create something that is competitive.

I’m guess that within 6 months Google will give up on coding and finally let their devs use Claude/Codex.

This isn’t a security problem, this is a GDM issue with GDM’s promises being far beyond their ability.

PunchTornado2• 3 months ago

Codex?

nine_k• 3 months ago

The point of dogfooding is exactly that: if we're unhappy, we're the ones to improve.

anthonypasq• 3 months ago

the engineers using gemini have no control over deepmind

DeathArrow3• 3 months ago

Are you in the Gemini team?

jensensbutton• 3 months ago

Note that coding is not the only use of Gemini or any of these models. It's also not what this article is talking about. Gemini can be not the best coding agent, but very good at other things.

j2kun4• 3 months ago

I for one can't tell the difference between Claude and Gemini for coding. And the internal agent tooling is many times faster than Claude Code in my experience.

jatora3• 3 months ago

Lie? Gemini CLI is unuseable. The IF of gemini models is atrocious. Honestly, how often does your gemini CLI go insane in thought loops and you have to stop it?

dekhn• 3 months ago

If you mean specifically the Gemini VS Code Extension: it's terrible compared to Claude Code or Codex. I don't know how they can get away with it. Just constant timeouts, weird failure modes, have to start a new chat to switch modes... but I don't think any of that is specific to gemini the model- it seems to be the extension.

As for actual solutions to problems ignoring the VS Code extension aspect, I find all three premiere models to be excellent coding agents for my purposes.

Groxx• 3 months ago

The overall quality of LLM coding tools is shockingly bad. I haven't found a single one without major issues, and many have the same problems reappear every few months, sometimes bad enough to almost break the entire thing (e.g. 100% failure rate in editing files, broken for weeks, with the same cause each time, multiple times in a year).

I'd say I'm surprised by it, but uh

llmslave• 3 months ago

they use claude code at deepmind

robohoe• 3 months ago

Antigravity comes to mind

kaueg2• 3 months ago

they use a web based vscode like (cider) with a custom agent

dzhiurgis2• 3 months ago

Not a Googler, but I use gemini in JetBrains Junie and have no issues with it. It's cheap, very fast and most importantly actually listens to you.

danpalmer3• 3 months ago

Yes. The models are good, the models are fast, and the internal tooling has caught up at this point too. There's a lot of UI/UX/tooling stuff that's still being worked through, integrations with VCS, and solving deeper problems that I probably can't talk about, but I'd say the frustrations of most are about the rate of change much more than the actual abilities.

One thing that's interesting is a bunch of internal thought leaders who swear by the Flash models over the Pro models. Whether this is true or not doesn't really matter, the interesting bit to me is that we are at a point with the models where "better" models are not necessarily more useful, and that faster with more work on the harnesses may be a better trade-off.

rapind2• 3 months ago

a bunch of internal thought leaders who swear by the Flash models over the Pro models

I'm coming around on this too. deepseek-v4-flash is impressive.

DeathArrow3• 3 months ago

One thing that's interesting is a bunch of internal thought leaders who swear by the Flash models over the Pro models.

I've seen people outside Google favoring flash Gemini models over the Pro.

There are also some benchmarks where flash models have higher scores, so yes, apparently speed does matter.

fractorial2• 3 months ago

You’re absolutely kidding yourself if you genuinely believe that.

goldenarm• 3 months ago

I'm quite surprised, 3 flash hallucinates way more than 3.1 pro, I stopped using it entirely. Unless you already have 3.2 flash internally :)

rebeccajae• 3 months ago

Most people that I have spoken to that are at Google have complained about the internal Gemini agent and seem to believe it's gotten significantly worse lately. Things like it completely forgetting how to call the tools and eventually giving up having wasted a bunch of time, or the agent completely ignoring code style guidance in an AGENTS.md-esque file.

My experience running Gemma 4 locally has been similar: after maybe one or two tool calls it starts making tool calls however it feels like. Just yesterday, I watched it redefine a tool like read_file(start, end) into read_file(start, number_of_bytes) and it refused to even consider that it was wrong.

brkn• 3 months ago

I would be interested to see how exactly the agent helped. How was it used, where did it lead to the given improvement and in how far would it have taken a human to come to the same solution.

j2kun4• 3 months ago

The blog post has many links to papers and preprints discussing this exact question.

Lt_Riza_Hawkeye• 3 months ago

The CANOS arxiv link says absolutely nothing about AlphaEvolve, Gemini, or LLMs. It seems to use purely traditional ML models. If AE did in fact write a quick script to test different configurations in order to optimize the results, they don't seem to have bothered to write about it.

I can't read the Nature paper about DeepConsensus, but from the summary, it doesn't really explain what role AE had in improving DC. It would be nice to be able to read about what role it actually played, and whether it used traditional or novel methods of performing it

armanj• 3 months ago

seems like `karpathy/autoresearch` on steroids

momojo3• 3 months ago

This reminds me of Antirez's "Don't fall into the anti-AI hype" [0]

In a sentence: These foundation models are really good at optimizing these extremely high level, extremely well defined problem spaces (ie multiply matrices faster). In Antirez's case, it's "make Redis faster".

There have been two reactions: "Oh it would never work for me" and "I have seen months of my life accomplished in an hour", and I think they're both right. I think we should be excited for Antirez, (who has since been popping off [1]), and I think the rest of us should rest easy knowing that LLM's can't (and maybe were never meant to) tackle the tacit-knowledge-filled, human-system-centric, ambiguously-defined-problem-space jobs most mortals work.

[0] https://antirez.com/news/158 [1] https://antirez.com/news/164

dinfinity4• 3 months ago

I think the rest of us should rest easy knowing that LLM's can't [...]

What if (when?) (AI-assisted) research moves AI beyond LLMs? Do you think that can't happen?

kubb5• 3 months ago

Not in the next decade. Won't get funded.

ActorNightly• 3 months ago

I mean, Google already has Mu Zero, which Im willing to bet has evolved quite a bit in private because if anything is going to get us closer to actual AI its that.

Realistically, one can build a AI capable of reasoning (i.e recurrent loops with branches) using very basic models that fit on a 3090, with multi agent configuration along the lines https://github.com/gastownhall/gastown. Nobody has done it yet because we don't know what the number of agents is required and what the prompts for those look like.

The fundamental philosophical problem is if that configuration is possible to arrive at using training, or do ai agents have to go through equivalent "evolution epocs" to be able to do all that in a simulated environment. Because in the case of those prompts and models, they have to be information agnostic.

poisonfountain4• 3 months ago

I think the rest of us should rest easy knowing that LLM's can't (and maybe were never meant to) tackle the tacit-knowledge-filled, human-system-centric, ambiguously-defined-problem-space jobs most mortals work

I don't believe that anymore, to be honest. Models are starting to get good at ambiguity. Claude Code now asks me when something is ambiguous. Soon, all meetings will be recorded, transcribed and stored in a well-indexed place for the agents to search when faced with ambiguity (free startup idea here!). If they can ask you now, they'll be able to search for the answers themselves once that's possible. In fact, they already do it now if you have a well-documented Notion/Confluence, it's just that nobody has.

It's probably harder to RL for "identify ambiguity" than RL'ing for performance algorithms, sure, but it's not impossible and it's in the works. It's just a matter of time now.

risyachka• 3 months ago

In coding the ambiguity is very, very limited and constrained compared to any non dev job that involves any decision making

TranquilMarmot• 3 months ago

Soon, all meetings will be recorded, transcribed and stored in a well-indexed place for the agents to search when faced with ambiguity (free startup idea here!)

We were doing that over at Vowel a few years back, unfortunately it didn't pan out because you're competing directly against Zoom, Google Meet, Microsoft Teams, etc. They are all (slowly) catching up to where we were as a scrappy startup 4 years ago.

It was truly game-changing to have all of your meetings in an easily searchable database. Even as a human.

Yokohiii8• 3 months ago

So self chosen total surveillance and transparency so your fav LLM can be better?

wuschel• 3 months ago

Unfortunately you can't record meetings in many jurisdictions, including court sessions. Hence we have to rely - for worse, or perhaps even for better - on human driven note taking.

momojo3• 3 months ago

Models are starting to get good at ambiguity

That's fair, and something I've observed too. I wish I had written "the rest of us shouldn't freak out and quit software today".

But here's another data point: At the biotech I work for, writing good code has never been the bottleneck. I actually told my boss that a paid Claude vs free subscription wouldn't be that much value because even if it took every piece of code or algorithm we've ever written and 10x-ed the hell out of them, we'd still be bottlenecked by the biology and physics which dictates that we wait 24 days for our histology assay pipeline.

I have a hunch most fields outside of software are this way. And I'm personally not planning to quit anytime soon.

dzhiurgis2• 3 months ago

Why record when it can build in realtime as meeting is going on.

Slack is kinda there with Salesforce - can do a lot already on Agentforce and in Slackbot, but two aren't integrated just yet and Slackbot doesn't support group chats/channels. One interesting aspect in this will be - who has superiority boss, client, analyst or developer?

mrdomino-3• 3 months ago

Tacit knowledge is definitionally not recorded in any of these systems. This proposes to solve the problem of tacit knowledge by getting rid of it. It is not clear to me if that solution is either possible or desirable.

cyanydeez4• 3 months ago

I'd say it's a malefactor of:

1. Amazing, you just tweaked 1% efficiency

2. You idiot, you just spent an hour trying to trouble shoot a hallucinated api.

On average, it's really hard to tell which ones going to win here.

dakolli• 3 months ago

Its not hard to tell at all, just look at how much it costs to run a 10T param model (especially with parallelized agents). Those costs are not worth the occasional slot machine-eque jackpot you get. For an entity like Google it might be worth it, but that's it. They definitely aren't going to let us use these things for cost they are now for much longer.

Imagine going back to 2020 and tell people in 6 years going to be able to spend $200.00 a month and be able to spin up $2mm in GPUs at full throttle to respond to your emails. None of this makes sense.

vonneumannstan2• 3 months ago

I think the rest of us should rest easy knowing that LLM's can't (and maybe were never meant to) tackle the tacit-knowledge-filled, human-system-centric, ambiguously-defined-problem-space jobs most mortals work.

A Statement all but guaranteed to look incredibly short sighted by 2030.

01100011• 3 months ago

The past few years has seen a great rise in casuals reminding us of AIs limitations only to be proven wrong in 6 months. I don't think we're close to AGI, but in 2 years I've gone from AI doubter to AI convert. It's not perfect, but I don't need it to be.

The real question to me is if the system can pay for itself. Economics are racing against efficiency gains and it's anyone's guess which wins.

wood_spirit• 3 months ago

I have found Claude et al good at quickly implementing the algorithm I have in mind effectively, as long as I ask lots of control questions and check code. They aren’t good at inventing non-mainstream algorithms though and often slip staggeringly short term shortcuts in though. They are still a tool and not yet the craftsman who wields tools effectively. This will steadily change, and the corners where the obscure algorithm wins will erode further too.

DoctorOetker2• 3 months ago

"... tackle the tacit-knowledge-filled, human-system-centric, ambiguously-defined-problem-space jobs most mortals work."

sounds like jobs involving legalese, politics, corruption and more generally involving pretending you don't understand something for which your income depends on not openly understanding something...

dandaka• 3 months ago

How many times we have to hear again about Erdös problems? :) It sounds like a great achievement for humanity at first, but after a while they keep coming back!

j2kun4• 3 months ago

There are only some 700 open Erdos problems left, so when they're all solved you can finally rest.

pilooch• 3 months ago

AlphaEvolve couples map-elites with LLMs. It's an key step in machine learning, in the vein of DQN for reinforcement learning.

AE brings diversity from the genetic algorithms community to large scale optmized deep learning and RL models.

It is a mandatory step for moving forward. The approach is clean and simple, while generic.

The only caveats is the per optimization problem definition of the map élites dimensions. But surely, this will get tackled somehow over the next few years.

If you don't know about map-elites, go look up Jean-Baptiste Mouret' s work and talks, it's both very interesting and universal.

stijntonk3• 3 months ago

I wish that Google would focus on bringing their Gemini 3.x models to GA, and provide enough capacity such that one not constantly has to fight with 429 errors.

It often feels like they do not want me to develop applications for corporate clients using their Vertex API. It is just such a shame, given that their models were so great for document analysis etc.

VladVladikoff• 3 months ago

Are you doing it on a free plan? I noticed they serve way more 429s on the free plan.

stijntonk3• 3 months ago

No, for clients we use paid Vertex AI accounts. We often need to host workloads in an EU region, which rules out “global” models (and probably better capacity).

In the past, we used a wrapper that round-robined across multiple projects to get enough quota. Luckily, many of our workloads are workflow-style tasks, so we can simply keep retrying on 429s.

Fun fact: for one of their services, I think it was Stitch, I noticed that my paid key kept hitting quota, while the free worked fine. That blew my mind.

AndrewKemendo• 3 months ago

From the comments it seems that this community (mostly career software people) is starting to move into a new phase of grief about the median software engineer losing their hoped for permanent place in society.

-2021-2024 was Denial

-2024-2025 was Anger and Bargaining

-2026 seems to be some combo of anger, bargaining and acceptance depending mostly on your class/age

artninja1988• 3 months ago

I think we are still in the denial phase.

nmitchko• 3 months ago

A fantastically simple solution to improving algorithms, I wish I had this years ago in activation engineering: https://blog.n.ichol.ai/llm-activation-engineering-an-easy-f...

How do I access AlphaEvolve?

Yokohiii8• 3 months ago

This is just a flex post. Be a billion dollar company or get out.

charleshn2• 3 months ago

They'll likely make it available at some point, but for now one can use OpenEvolve [0] which is not quite as good but should be a good start to use the same LLM-driven evolutionary framework.

[0] https://github.com/algorithmicsuperintelligence/openevolve

arian_• 3 months ago

We went from 'AI will replace programmers' to 'AI will help programmers' to 'AI writes code while other AI reviews it' in about 18 months. At this rate the humans are just providing the electricity.

svieira• 3 months ago

In advertising and marketing, WPP used AlphaEvolve to refine AI model components, navigating complex, high-dimensional campaign data and achieving 10% accuracy gains over their competitive manual model optimizations.

Ah good, we're getting closer and closer to Venus, Inc. every day. /s

agluszak• 3 months ago

Meanwhile Gemini CLI has been broken for months!

https://github.com/google-gemini/gemini-cli/issues/22141

guybedo• 3 months ago

and yet Gemini still can't code

kaueg2• 3 months ago

Welcome to HN @berlianta; TIL green username === new user in HN; Stories posted by new users are called noobstories [1];

[1] https://news.ycombinator.com/noobstories

berlianta• 2 months ago

No need for a welcome message, just stick to the topic. Thank you

ainch• 3 months ago

The AI CEOs love to pontificate about AI curing cancer, but it seems like DeepMind is the only one actively working on these research problems, while OpenAI/Anthropic largely chase enterprise/coding revenue.

rapind2• 3 months ago

Google can self fund from their war chest while OpenAI and Anthropic are hat in hand.

tim333• 3 months ago

Also I think Hassabis was long focussed on using AI for science. He got a PhD in neuroscience before starting DeepMind.

igorpcosta• 3 months ago

There's not a lot of opportunities in this space yet. This is the closest we can get to High degree solver kinda of problem.

There are only 3 companies doing this to date: Google, Sakana AI and Autohand AI.

cpard• 3 months ago

All the *Evolve publications have very impressive results but from the time I’ve spent on the information published I feel that the attention goes to the LLMs and the AI side of things, although the outcomes reported are in almost all cases the result of very well designed environments for both the LLM and the evolutionary algorithm to work well.

This paper here is a great example of that and it’s worth a reading.

Magellan: Autonomous Discovery of Novel Compiler Optimization Heuristics with AlphaEvolve https://arxiv.org/abs/2601.21096

sjhalani7• 3 months ago

This is crazy- the fact that it is helping with stuff like quantum too is huge!

zkmon• 3 months ago

An issue I have been noticing with claude is, for simple tasks, it gives extremely bloated code and artifacts, which sometime does not even work. Gemini balances it quite well, by giving a working solution with the exact amount of code and minimal complexity, that is easier to manage.

The only thing I go to Claude these days, is for front-end code (HTML). Here also, it gives too much of CSS code (60% of the file size), but I'm OK with that as it gives a bit of polished look, though heavy on file size.

AlphaEvolve: Gemini-powered coding agent scaling impact across fields

Related Stories