/AI6h ago

DeepSeek's Deli Chen releases Deli_AutoResearch, a tool that autonomously generated three complete AI survey papers

The agent-authored papers averaged an 8.5/10 peer-review score.

--0--
#421Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞)
Deli Chen
Original post
Deli Chen@victor207755822#1540inAI

Happy to share my research vision — opinions are all my own. Here's why I'm writing these three surveys:

AutoResearch: a productivity tool that helps researchers validate ideas faster, so they can focus more on thinking itself. 🧠

Self-iteration: a clear signal we're seeing — model improvement is no longer a one-shot deal. It requires continuous, proactive information gathering and parameter updates. 🔄

Long-horizon tasks: current models already handle most simple problems, but real-world production problems are deeply complex and demand long-cycle, multi-step decisions. Solving this is what truly pushes AGI forward. 🧭

So the first survey is about an efficiency tool, the second points to a clear direction that needs improvement, and the third tackles long-horizon complexity.

These surveys will be continuously updated. Looking forward to exchanging ideas with everyone. See you at the summit of AGI. ⛰️

Deli Chen@victor207755822

🔥 Deli_AutoResearch Project just Updated~!

Three complete survey papers (one brand-new, two updated), entirely authored by LLM agents. Not a single human-written paragraph. 🤖📄

📊 By the numbers: • 941 references across the papers • 190 pages total • Avg peer review score: 8.5/10 (up from 6.0 after 14 rounds of AI-driven revision) • Runtime: ~38 hours

🆕 New Paper #3: Long-Horizon Sequential Decision-Making 57 pages · 384 references 🔗 https://victorchen96.github.io/auto_research/paper.html

🧠 Open-source: Scientific Paper Writing SKILL 🔗 https://victorchen96.github.io/auto_research/skill/paper-writing.html

🔮 What’s next? The bottleneck is no longer writing quality — it’s research taste.

Next is targeting hypothesis generation & novelty detection for fully original AI research. 🚀

#AIResearch #LLMAgents #AutoResearch #OpenSourceAI #AGIForEveryone #DeepSeek #GPT-Image2

(Disclaimer: All opinions are my own and do not represent the views of any company or organization.)

10:34 AM · Jun 4, 2026 · 1.7K Views
Sentiment
Sentiment building, check back later.
Cluster Engagement
-
Views
-
Comments
-
Reposts
-
Bookmarks
Expand data
Posts from X
Most Activity
Most Activity
VIEWS2.6KBOOKMARKS13LIKES19REPLIES4
Deli Chen@victor207755822

Using my own research paper-writing ability as the baseline, here's how I'd rate the current version of Deli AutoResearch:

Literature collection — 150% of my ability. It can auto-fetch all papers, read fast, and reliably pick out the important ones. ⭐⭐⭐

Research insight extraction & logical structuring — about 50%. It mostly surfaces mainstream views and offers only limited novelty. ⭐

Experiment design — about 70%. For well-established benchmarks, it designs effective experiments. But for genuinely innovative work, it still can't pull it off. ⭐⭐

Deli Chen@victor207755822

🔥 Deli_AutoResearch Project just Updated~!

Three complete survey papers (one brand-new, two updated), entirely authored by LLM agents. Not a single human-written paragraph. 🤖📄

📊 By the numbers: • 941 references across the papers • 190 pages total • Avg peer review score: 8.5/10 (up from 6.0 after 14 rounds of AI-driven revision) • Runtime: ~38 hours

🆕 New Paper #3: Long-Horizon Sequential Decision-Making 57 pages · 384 references 🔗 https://victorchen96.github.io/auto_research/paper.html

🧠 Open-source: Scientific Paper Writing SKILL 🔗 https://victorchen96.github.io/auto_research/skill/paper-writing.html

🔮 What’s next? The bottleneck is no longer writing quality — it’s research taste.

Next is targeting hypothesis generation & novelty detection for fully original AI research. 🚀

#AIResearch #LLMAgents #AutoResearch #OpenSourceAI #AGIForEveryone #DeepSeek #GPT-Image2

(Disclaimer: All opinions are my own and do not represent the views of any company or organization.)

6hViews 2.6KLikes 19Bookmarks 13
RETWEETS2
Deli Chen@victor207755822

Using my own research paper-writing ability as the baseline, here's how I'd rate the current version of Deli AutoResearch:

Literature collection — 150% of my ability. It can auto-fetch all papers, read fast, and reliably pick out the important ones. ⭐⭐⭐

Research insight extraction & logical structuring — about 50%. It mostly surfaces mainstream views and offers only limited novelty. ⭐

Experiment design — about 70%. For well-established benchmarks, it designs effective experiments. But for genuinely innovative work, it still can't pull it off. ⭐⭐

Deli Chen@victor207755822

🔥 Deli_AutoResearch Project just Updated~!

Three complete survey papers (one brand-new, two updated), entirely authored by LLM agents. Not a single human-written paragraph. 🤖📄

📊 By the numbers: • 941 references across the papers • 190 pages total • Avg peer review score: 8.5/10 (up from 6.0 after 14 rounds of AI-driven revision) • Runtime: ~38 hours

🆕 New Paper #3: Long-Horizon Sequential Decision-Making 57 pages · 384 references 🔗 https://victorchen96.github.io/auto_research/paper.html

🧠 Open-source: Scientific Paper Writing SKILL 🔗 https://victorchen96.github.io/auto_research/skill/paper-writing.html

🔮 What’s next? The bottleneck is no longer writing quality — it’s research taste.

Next is targeting hypothesis generation & novelty detection for fully original AI research. 🚀

#AIResearch #LLMAgents #AutoResearch #OpenSourceAI #AGIForEveryone #DeepSeek #GPT-Image2

(Disclaimer: All opinions are my own and do not represent the views of any company or organization.)

6hViews 2.6KLikes 19Bookmarks 13