DeepSeek researcher Deli Chen publishes an autonomous research agent taxonomy featuring a fictional Level 5 system

DeepSeek researcher Deli Chen publishes an autonomous research agent taxonomy featuring a fictional Level 5 system · Digg

Posts from X

Most Activity

VIEWS15.6KBOOKMARKS21LIKES66RETWEETS4REPLIES16

Deli Chen@victor207755822

With CodeAgent, I can finally pick up so many things I’d dropped due to low energy. Blogging is one of them. This blog is ~1% me, 99% the agent 😂 https://victorchen96.github.io/auto_research_survey.pdf (Disclaimer: Just doing this for fun and testing my skill: DeliAutoResearch😆. Not a rigorous academic paperc. All opinions are my own and do not represent the views of any company or organization.)

Deli Chen@victor207755822

Hey everyone！ after neglecting it for ages, I’m finally updating my site just for fun~ (things got so much more efficient with a code agent 😂) Come check it out! 🚀

https://victorchen96.github.io/

34d15.6K6621

Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞)@teortaxesTex

This is essentially a joke, but I think the intention is serious. In time, DeepSeek will do more than "Deli's AutoResearch skill.md".

Deli Chen@victor207755822

34d5.9K2812

Deli Chen@victor207755822

@Radipdegen you are right hhh， my idea is the 1% 😆

34d2182

big goose@Anonyous_FPS

@victor207755822 ？我才发现是陈德里。我之前都是默念的陈大力🙃，非常抱歉，看来我的注意力需要改善一下

34d781

Deli Chen@victor207755822

Production stats: Paper 6 iterations total (V1:4, V2:1, V3:1). V1 first draft took 76 min. Total wall clock: 6 days. ~108 agent rounds, ~648K tokens. 2,234 lines of LaTeX. 103 references, all verified. 45 → 46 pages. 7 figures + 4 tables. Paper is now 46 pages, 538KB. 😂

Personal hot take: Code Agent is causing crazy inflation in CS papers — the same work used to take at least a month. 📈🤯

34d49

Deli Chen@victor207755822

My brain's total CPU time on this process: less than 2 hours. 🧠⏱️

34d1671

adhdsloth@adhdsloth150217

@victor207755822 That's pretty cool actually. Bit curious about the "DeliAutoResearch" skill now

34d164

flyingpetals@flyingpetal472

@victor207755822 DeepSeek 能出点自带的 auto research 的框架或者训练agent swarm能力吗，角色扮演展现出的multi agents系统可行性高很多，这波harness 要出就搞个大的。

34d351

Eri.ck@erickxie

@victor207755822 It's now called the 2nd brain 🧠 creation right?

BTW the Skill needs some Spelling Check function before posting. 😅

34d96

Winston B.@DoDataThings

@teortaxesTex Reads like the early-Claude-Code give-it-bash-and-trust phase, where one skill.md ran half the workflow. The serious version of that joke is a lab that ships SKILL.md files weekly and trains the next checkpoint on what worked.

34d77

Strata@ChainZenit

@victor207755822 oh nice, agent made the updates way smoother

34d52

Alex UGift@Radipdegen

@victor207755822 building your own site is a rite of passage we all skip til the guilt kicks in

using a code agent is cheating but smart

34d46

unbug@unbug

@victor207755822 Step further, Try AI for daily things you’ll be blow mind.

34d41

0xnb@neuralbroker

@victor207755822 Bro having fun on X :) Just a small tweak , redesign it in a clean research-style template, remove the AI-slop emojis/content, change the username to delichen, and make it feel more friendly.

DeloGOD :)

34d24

Alok Bishoyi@alokbishoyi97

@victor207755822 have been myself running a custom loop sinialr to this via evo to go through papers ( contextually relvant ones ) and implement and commit / discard based on the rules

http://Github.com/evo-hq/evo

34d16

Alex UGift@Radipdegen

@victor207755822 That’s 1% is what make us valuable

34d11

John Shina@JohnShina91

@victor207755822 Impressive resume there.

34d11

flyingpetals@flyingpetal472

@Anonyous_FPS @victor207755822 他在小红书上都有号的🤗😂

34d7

Strata@ChainZenit

@victor207755822 this is actually pretty cool ngl

34d7

HammerChain@anaphor1c

@victor207755822 v4 pro or flash?

34d2