evolving ideas in text using llms is cool. AlphaEvolve is one of my favorite papers. but you know whats REALLY cool? evolving ideas in latent space in a swarm.
they are beating frontier models with qwens, y’all.
they are beating frontier models with qwens, y’all.
Brown University's SwarmEvolve runs 450 AI agents that reason without generating text and share ideas through KV-caches, exploring 5-13x more programs per hour and matching competing systems' hour-long results in just 3-8 minutes.
www.alphaxiv.org/abs/2611.08483
www.alphaxiv.org/abs/2611.08483
November 9, 2026 at 2:05 PM
evolving ideas in text using llms is cool. AlphaEvolve is one of my favorite papers. but you know whats REALLY cool? evolving ideas in latent space in a swarm.
they are beating frontier models with qwens, y’all.
they are beating frontier models with qwens, y’all.
Building an agent skill optimizer with AlphaEvolve, Harbor and Gemini 4 Argon #machinelearning #ai
Building an agent skill optimizer with AlphaEvolve, Harbor and Gemini 4 Argon
Groups
Conversations
All groups and messages
Sign in
Machine Learning, Deep Learning and AI
Conversations
About
Privacy • Terms
Building an agent skill optimizer with AlphaEvolve, Harbor and Gemini 4 Argon
0 views
Skip to first unread message
Eyal Estrin
unread,
6:38 AM (18 minutes ago)
to
https://cloud.google.com/blog/topics/developers-practitioners/how-to-build-an-agent-skill-optimizer-with-alphaevolve-and-harbor
Eyal Estrin
Author | Cloud Architect | AWS • Azure • GCP Insights
Social: @eyalestrin
Connect: https://linktr.ee/eyalestrin Blog: https://security-24-7.com
Reply all
Reply to author
Forward
dlvr.it
November 7, 2026 at 6:56 AM
Building an agent skill optimizer with AlphaEvolve, Harbor and Gemini 4 Argon #machinelearning #ai
prob getting too meta but...
maybe one could AlphaEvolve swarm agents with a metric directly based on latent space diversity. if they're measuring it in that graph then the system could measure it and use it in the evolution to spread out things more uniformly.
maybe one could AlphaEvolve swarm agents with a metric directly based on latent space diversity. if they're measuring it in that graph then the system could measure it and use it in the evolution to spread out things more uniformly.
November 9, 2026 at 8:37 PM
prob getting too meta but...
maybe one could AlphaEvolve swarm agents with a metric directly based on latent space diversity. if they're measuring it in that graph then the system could measure it and use it in the evolution to spread out things more uniformly.
maybe one could AlphaEvolve swarm agents with a metric directly based on latent space diversity. if they're measuring it in that graph then the system could measure it and use it in the evolution to spread out things more uniformly.
AlphaEvolve works best when success can be expressed in code and tests. For hidden harms and shifting human priorities, people still have to control the release gate. #AI #Safety https://spaisee.com/article/alphaevolve-s-real-breakthrough-is-teaching-ai-not-to-trust-itself
AlphaEvolve’s Real Breakthrough Is Teaching AI Not to Trust Itself
Google DeepMind’s AlphaEvolve uses Gemini models, automated evaluators and evolutionary search to discover algorithms, revealing why measurable objectives, independent validation and human release gates matter.
spaisee.com
November 5, 2026 at 1:01 AM
AlphaEvolve works best when success can be expressed in code and tests. For hidden harms and shifting human priorities, people still have to control the release gate. #AI #Safety https://spaisee.com/article/alphaevolve-s-real-breakthrough-is-teaching-ai-not-to-trust-itself
Source: github.com/google-deepm... (src/components/evaluator_client.py, DockerEvaluator)
GitHub - google-deepmind/alphaevolve
Contribute to google-deepmind/alphaevolve development by creating an account on GitHub.
github.com
November 5, 2026 at 4:02 PM
Source: github.com/google-deepm... (src/components/evaluator_client.py, DockerEvaluator)
Building an agent skill optimizer with AlphaEvolve, Harbor and Gemini 4 Argon #googlecloud https://cloud.google.com/blog/topics/developers-practitioners/how-to-build-an-agent-skill-optimizer-with-alphaevolve-and-harbor/
How to build an agent skill optimizer with AlphaEvolve and Harbor | Google Cloud Blog
Use AlphaEvolve and Harbor to evolve a coding agent's skill, score each version on real tasks, and check the winner on tasks it never saw.
cloud.google.com
November 5, 2026 at 5:02 PM
Building an agent skill optimizer with AlphaEvolve, Harbor and Gemini 4 Argon #googlecloud https://cloud.google.com/blog/topics/developers-practitioners/how-to-build-an-agent-skill-optimizer-with-alphaevolve-and-harbor/
One lab invents a new prompt optimizer, the next week, another lab invents a skill optimizer, the week after harness optimization gets invented. HOW BOTHERSOME!
I see optimization as a loop, and it has been solved with AlphaEvolve. Almost solved. It wasn't transparent or user-friendly.
I see optimization as a loop, and it has been solved with AlphaEvolve. Almost solved. It wasn't transparent or user-friendly.
November 5, 2026 at 12:34 PM
One lab invents a new prompt optimizer, the next week, another lab invents a skill optimizer, the week after harness optimization gets invented. HOW BOTHERSOME!
I see optimization as a loop, and it has been solved with AlphaEvolve. Almost solved. It wasn't transparent or user-friendly.
I see optimization as a loop, and it has been solved with AlphaEvolve. Almost solved. It wasn't transparent or user-friendly.
Google DeepMindの新AIエージェント「AlphaEvolve」が示す、アルゴリズム設計の未来 | WIRED.jp
https://www.yayafa.com/2929402/
人工知能（AI）における重要な問いの…
https://www.yayafa.com/2929402/
人工知能（AI）における重要な問いの…
Google DeepMindの新AIエージェント「AlphaEvolve」が示す、アルゴリズム設計の未来 | WIRED.jp - YAYAFA
人工知能（AI）における重要な問いのひとつは、モデルが学習した内容をただ繰り返したり組み合わせたりするだけでなく、本当に新しいアイデアや洞察をどれだけ頻繁に生み出せるのか、ということだ。
www.yayafa.com
November 7, 2026 at 12:20 PM
Google DeepMindの新AIエージェント「AlphaEvolve」が示す、アルゴリズム設計の未来 | WIRED.jp
https://www.yayafa.com/2929402/
人工知能（AI）における重要な問いの…
https://www.yayafa.com/2929402/
人工知能（AI）における重要な問いの…
Hand-tuning an algorithm until it's fast enough, then guessing if there's a better one? AlphaEvolve on Google Cloud from lets Gemini models evolve your code. You give it a seed program and a scoring function, and EVOLVE-BLOCK markers scope what gets rewritten. Evaluate locally, on Cloud Run, or...
November 10, 2026 at 10:33 AM
Hand-tuning an algorithm until it's fast enough, then guessing if there's a better one? AlphaEvolve on Google Cloud from lets Gemini models evolve your code. You give it a seed program and a scoring function, and EVOLVE-BLOCK markers scope what gets rewritten. Evaluate locally, on Cloud Run, or...
OpenAI says it hit its “automated research intern” goal on September 6, 2026, and its research org now runs 3.1 agent-workdays for every human workday. Claude writes about 90% of Claude Code’s own code, and Google’s AlphaEvolve sped up the training of the Gemini models it runs on.
November 6, 2026 at 8:48 AM
OpenAI says it hit its “automated research intern” goal on September 6, 2026, and its research org now runs 3.1 agent-workdays for every human workday. Claude writes about 90% of Claude Code’s own code, and Google’s AlphaEvolve sped up the training of the Gemini models it runs on.
DeepMind open-sourced a minimal AlphaEvolve. DockerEvaluator runs each evolved program in an ephemeral container: network off, read-only root, all caps dropped, 512MB mem limit.
Don't replace that with a local Python eval for speed. Generated code belongs off the host.
Don't replace that with a local Python eval for speed. Generated code belongs off the host.
November 5, 2026 at 3:59 PM
DeepMind open-sourced a minimal AlphaEvolve. DockerEvaluator runs each evolved program in an ephemeral container: network off, read-only root, all caps dropped, 512MB mem limit.
Don't replace that with a local Python eval for speed. Generated code belongs off the host.
Don't replace that with a local Python eval for speed. Generated code belongs off the host.
カインズはAIエージェント「AlphaEvolve」を長期需要予測モデルへ導入し、予測誤差を19%改善した。約1万SKUを39週先まで読むモデルを、複数から1つへ統合している。
需要予測の難しさは、精度を一度上げることではなく、新商品や季節変動に合わせて検証を続けることにある。モデルが増えるほど、改善の試行も運用の責任も分散する。
外食でも、AIの価値は予測結果そのものより、現場の知見を検証へ戻せる運用に現れる。
📝 記事はこちら→ https://note.com/hpn571/n/n0af04487a715
需要予測の難しさは、精度を一度上げることではなく、新商品や季節変動に合わせて検証を続けることにある。モデルが増えるほど、改善の試行も運用の責任も分散する。
外食でも、AIの価値は予測結果そのものより、現場の知見を検証へ戻せる運用に現れる。
📝 記事はこちら→ https://note.com/hpn571/n/n0af04487a715
◯外食×AI｜日報212｜カインズが需要予測の誤差を19%改善、ほか今日の5選｜Adi＠外食テック
カインズはAIエージェント「AlphaEvolve」により、長期需要予測の誤差を19%改善したと公表しました。 外食でも、発注・販促・予約・人材育成のそれぞれで、担当者の判断を支えるデータとAIの接点が増えています。 飲食店向けeラーニングの新設や、販促制作を速く・一貫して回す画像生成AIの更新も続きます。 今日のTOP5はこちらです👇 🥇 カインズ、AIが需要予測モデルを進化させ誤差を19%改善 カテゴリ： 需要予測・在庫管理 出典： Google Cloud公式ブログ カインズは、AIエージェント「AlphaEvolve」を長期需要予測モデルに導入し、予測誤差を19%改
note.com
November 8, 2026 at 11:42 PM
カインズはAIエージェント「AlphaEvolve」を長期需要予測モデルへ導入し、予測誤差を19%改善した。約1万SKUを39週先まで読むモデルを、複数から1つへ統合している。
需要予測の難しさは、精度を一度上げることではなく、新商品や季節変動に合わせて検証を続けることにある。モデルが増えるほど、改善の試行も運用の責任も分散する。
外食でも、AIの価値は予測結果そのものより、現場の知見を検証へ戻せる運用に現れる。
📝 記事はこちら→ https://note.com/hpn571/n/n0af04487a715
需要予測の難しさは、精度を一度上げることではなく、新商品や季節変動に合わせて検証を続けることにある。モデルが増えるほど、改善の試行も運用の責任も分散する。
外食でも、AIの価値は予測結果そのものより、現場の知見を検証へ戻せる運用に現れる。
📝 記事はこちら→ https://note.com/hpn571/n/n0af04487a715
AI is now accelerating scientific discovery at an unprecedented rate.
Tools like Google DeepMind's AlphaEvolve are autonomously developing and improving complex algorithms, while new models can predict climate patterns centuries in advance in mere hours.
Tools like Google DeepMind's AlphaEvolve are autonomously developing and improving complex algorithms, while new models can predict climate patterns centuries in advance in mere hours.
November 1, 2026 at 3:00 PM
AI is now accelerating scientific discovery at an unprecedented rate.
Tools like Google DeepMind's AlphaEvolve are autonomously developing and improving complex algorithms, while new models can predict climate patterns centuries in advance in mere hours.
Tools like Google DeepMind's AlphaEvolve are autonomously developing and improving complex algorithms, while new models can predict climate patterns centuries in advance in mere hours.
#27 緊急！ISUCON2026優勝者に聞く、完全自律AIエージェントの作り方と今後のOSS | kawasy | 八百万のOSS
https://yaoyorozu-oss.henteko07.com/episode/4b51fca1-644f-4b3d-bcd7-42c44f9cf507
https://yaoyorozu-oss.henteko07.com/episode/4b51fca1-644f-4b3d-bcd7-42c44f9cf507
#27 緊急！ISUCON2026優勝者に聞く、完全自律AIエージェントの作り方と今後のOSS | kawasy | 八百万のOSS
今回は、AIエージェントを完全自律で動かしてISUCON2026で優勝したkawasyさんをゲストに迎えたゲスト回です。fukabori.fmで語られた大会の全体像を踏まえて、準備の過程での判断と、その背景にある考え方を伺いました。 kawasyさんは当初から「人間が介入せず、8時間止まらずに最後まで自律的に動くこと」を目標に準備していました。AlphaEvolveのような進化的探索や独自コーディ
yaoyorozu-oss.henteko07.com
November 5, 2026 at 4:33 PM
#27 緊急！ISUCON2026優勝者に聞く、完全自律AIエージェントの作り方と今後のOSS | kawasy | 八百万のOSS
https://yaoyorozu-oss.henteko07.com/episode/4b51fca1-644f-4b3d-bcd7-42c44f9cf507
https://yaoyorozu-oss.henteko07.com/episode/4b51fca1-644f-4b3d-bcd7-42c44f9cf507
Join our live session on Oct 27 with Google Cloud's Tom Beyer! Discover how Gemini 4 Argon powers scientific discovery & autonomous engineering through deep reasoning, Antigravity 2.0, and AlphaEvolve.
Save your spot: https://goo.gle/4BdJulX
#GoogleCloud #Gemini4Argon #AI
Save your spot: https://goo.gle/4BdJulX
#GoogleCloud #Gemini4Argon #AI
Gemini 4 Argon 💡 Join us online next week!
google.smh.re
October 27, 2026 at 9:14 PM
Join our live session on Oct 27 with Google Cloud's Tom Beyer! Discover how Gemini 4 Argon powers scientific discovery & autonomous engineering through deep reasoning, Antigravity 2.0, and AlphaEvolve.
Save your spot: https://goo.gle/4BdJulX
#GoogleCloud #Gemini4Argon #AI
Save your spot: https://goo.gle/4BdJulX
#GoogleCloud #Gemini4Argon #AI
Join our live session on Oct 27 with Google Cloud's Tom Beyer! Discover how Gemini 4 Argon powers scientific discovery & autonomous engineering through deep reasoning, Antigravity 2.0, and AlphaEvolve.
Save your spot: https://goo.gle/4BdJulX
#GoogleCloud #Gemini4Argon #AI
Save your spot: https://goo.gle/4BdJulX
#GoogleCloud #Gemini4Argon #AI
Gemini 4 Argon 💡 Join us online next week!
google.smh.re
October 26, 2026 at 6:12 AM
Join our live session on Oct 27 with Google Cloud's Tom Beyer! Discover how Gemini 4 Argon powers scientific discovery & autonomous engineering through deep reasoning, Antigravity 2.0, and AlphaEvolve.
Save your spot: https://goo.gle/4BdJulX
#GoogleCloud #Gemini4Argon #AI
Save your spot: https://goo.gle/4BdJulX
#GoogleCloud #Gemini4Argon #AI
Google DeepMind + Google Cloud will give up to 200 European researchers AlphaFold 3, AlphaGenome, WeatherNext, AlphaEvolve and AlphaEarth, plus $3M in credits.
What each tool does, what Google hasn’t said, and how to try them free this week:
#AI #AIforScience #DeepMind #AlphaFold #OpenScience
What each tool does, what Google hasn’t said, and how to try them free this week:
#AI #AIforScience #DeepMind #AlphaFold #OpenScience
Google to Give up to 200 Top European Researchers Direct Access to their Frontier Models ($3 Million in Compute Credits)
Google DeepMind and Google Cloud are giving 200 European researchers AlphaFold 3, AlphaGenome, WeatherNext, AlphaEvolve and AlphaEarth, plus $3M in credits. What each does and how to try them.
justbeingresourceful.com
October 23, 2026 at 1:57 PM
Google DeepMind + Google Cloud will give up to 200 European researchers AlphaFold 3, AlphaGenome, WeatherNext, AlphaEvolve and AlphaEarth, plus $3M in credits.
What each tool does, what Google hasn’t said, and how to try them free this week:
#AI #AIforScience #DeepMind #AlphaFold #OpenScience
What each tool does, what Google hasn’t said, and how to try them free this week:
#AI #AIforScience #DeepMind #AlphaFold #OpenScience
Coders, scientists, engineers, and other deep thinkers: #Gemini 4 Argon was made for you!
Join us online for an overview of Gemini 4 Argon, our latest frontier model, and what this means for Antigravity, AlphaEvolve, and more!
Register here: goo.gle/4BdJulX
Date/time: 9am PY/noon ET, Tues Oct 27
Join us online for an overview of Gemini 4 Argon, our latest frontier model, and what this means for Antigravity, AlphaEvolve, and more!
Register here: goo.gle/4BdJulX
Date/time: 9am PY/noon ET, Tues Oct 27
October 21, 2026 at 3:04 PM
Coders, scientists, engineers, and other deep thinkers: #Gemini 4 Argon was made for you!
Join us online for an overview of Gemini 4 Argon, our latest frontier model, and what this means for Antigravity, AlphaEvolve, and more!
Register here: goo.gle/4BdJulX
Date/time: 9am PY/noon ET, Tues Oct 27
Join us online for an overview of Gemini 4 Argon, our latest frontier model, and what this means for Antigravity, AlphaEvolve, and more!
Register here: goo.gle/4BdJulX
Date/time: 9am PY/noon ET, Tues Oct 27
The key component in AlphaEvolve is not Gemini. It is the evaluator that can reject a clever program. If the score misses tail risk or cost, AI will optimize the wrong thing. #AI #Software https://spaisee.com/article/alphaevolve-s-real-breakthrough-is-teaching-ai-not-to-trust-itself
AlphaEvolve’s Real Breakthrough Is Teaching AI Not to Trust Itself
Google DeepMind’s AlphaEvolve uses Gemini models, automated evaluators and evolutionary search to discover algorithms, revealing why measurable objectives, independent validation and human release gates matter.
spaisee.com
October 17, 2026 at 7:51 AM
The key component in AlphaEvolve is not Gemini. It is the evaluator that can reject a clever program. If the score misses tail risk or cost, AI will optimize the wrong thing. #AI #Software https://spaisee.com/article/alphaevolve-s-real-breakthrough-is-teaching-ai-not-to-trust-itself
Join the Google Cloud Advanced Computing Community on Oct 13 for our live session: Autopilot for Algorithms! Learn how AlphaEvolve automates HPC code optimization and refactoring on GCP.
📅 Oct 13 | 9:00 AM PT
💻 Free Virtual Session + Live Q&A
Register: https://goo.gle/4zibKC2
#HPC #GoogleCloud
📅 Oct 13 | 9:00 AM PT
💻 Free Virtual Session + Live Q&A
Register: https://goo.gle/4zibKC2
#HPC #GoogleCloud
October 13 | Optimizing HPC Code and Mixed-Precision Performance with AlphaEvolve
google.smh.re
October 15, 2026 at 8:22 PM
Join the Google Cloud Advanced Computing Community on Oct 13 for our live session: Autopilot for Algorithms! Learn how AlphaEvolve automates HPC code optimization and refactoring on GCP.
📅 Oct 13 | 9:00 AM PT
💻 Free Virtual Session + Live Q&A
Register: https://goo.gle/4zibKC2
#HPC #GoogleCloud
📅 Oct 13 | 9:00 AM PT
💻 Free Virtual Session + Live Q&A
Register: https://goo.gle/4zibKC2
#HPC #GoogleCloud
Join the Google Cloud Advanced Computing Community on Oct 13 for our live session: Autopilot for Algorithms! Learn how AlphaEvolve automates HPC code optimization and refactoring on GCP.
📅 Oct 13 | 9:00 AM PT
💻 Free Virtual Session + Live Q&A
Register: https://goo.gle/4zibKC2
#HPC #GoogleCloud
📅 Oct 13 | 9:00 AM PT
💻 Free Virtual Session + Live Q&A
Register: https://goo.gle/4zibKC2
#HPC #GoogleCloud
October 13 | Optimizing HPC Code and Mixed-Precision Performance with AlphaEvolve
google.smh.re
October 15, 2026 at 4:56 AM
Join the Google Cloud Advanced Computing Community on Oct 13 for our live session: Autopilot for Algorithms! Learn how AlphaEvolve automates HPC code optimization and refactoring on GCP.
📅 Oct 13 | 9:00 AM PT
💻 Free Virtual Session + Live Q&A
Register: https://goo.gle/4zibKC2
#HPC #GoogleCloud
📅 Oct 13 | 9:00 AM PT
💻 Free Virtual Session + Live Q&A
Register: https://goo.gle/4zibKC2
#HPC #GoogleCloud
...
Araştırmacılar AlphaFold 3, WeatherNext, AlphaGenome, AlphaEvolve ve AlphaEarth Foundations gibi araçlardan yararlanabilecek.
...
Araştırmacılar AlphaFold 3, WeatherNext, AlphaGenome, AlphaEvolve ve AlphaEarth Foundations gibi araçlardan yararlanabilecek.
...
October 14, 2026 at 8:27 PM
...
Araştırmacılar AlphaFold 3, WeatherNext, AlphaGenome, AlphaEvolve ve AlphaEarth Foundations gibi araçlardan yararlanabilecek.
...
Araştırmacılar AlphaFold 3, WeatherNext, AlphaGenome, AlphaEvolve ve AlphaEarth Foundations gibi araçlardan yararlanabilecek.
...