kimjune01

kimjune01

15 日前議論

lean compiles or it doesnt

kimjune01

16 日前議論

this is known as the oracle problem

kimjune01

22 日前議論

one of those words that lost meaning, like samaritan or literally

kimjune01

先月議論

In the days when Sussman was a novice, Minsky once came to him as he sat hacking at the PDP-6.

"What are you doing?", asked Minsky.

"I am training a randomly wired neural net to play Tic-tac-toe", Sussman replied.

"Why is the net wired randomly?", asked Minsky.

"I do not want it to have any preconceptions of how to play", Sussman said.

Minsky then shut his eyes.

"Why do you close your eyes?" Sussman asked his teacher.

"So that the room will be empty."

At that moment, Sussman was enlightened.

kimjune01

先月議論

oh this again, i found this on stumbleupon circa 2007

kimjune01

2 か月前議論

using AI to red-team your thoughts and assumptions is the fastest way to get smart since the dawn of time

kimjune01

2 か月前議論

coopetition

kimjune01

2 か月前議論

numpy is the APL reincarnate

kimjune01

2 か月前議論

I appealed one time, took 3 weeks to get my 30 dollars back. will never deal with that again

kimjune01

2 か月前議論

I've been running a public experiment on how far I can take agentic coding harnesses. When I ran out of my own stuff to work on, I looked to open source.

So far, I got the harness to create 368 PRs, 110 of which were resolved. Of those, I got a 56% merge rate.

The harness: https://github.com/kimjune01/sweep

But as soon as I realized what I was doing to maintainers, I worked all night to build the same pipe but backwards for PR review.

The defense: https://github.com/kimjune01/immune

Which means: bots on the contributor side, bots on the maintainer side. The surprise? What's left is two humans on either end. I accidentally invented a H2H protocol.

I'm happy to share what I learned.

kimjune01

2 か月前議論

Although Arena is adversarial and resistant to goodharting, it's not immune. Models that train on Arena converge on helpfulness, not necessarily truthiness

kimjune01

2 か月前議論

are merged PRs a measure of velocity? github.com/kimjune01/

kimjune01

2 か月前議論

we lack the shared data structure for it

kimjune01

2 か月前議論

the garbage gets smarter with each iteration

kimjune01

3 か月前議論

AI labs should compete on a bench that's adversarial, such as go or Starcraft

kimjune01

3 か月前議論

internet ads are largely a story of parasitic exchanges using information asymmetry to suck as much surplus out of the middle as possible.

https://june.kim/advertising-journey/

kimjune01

3 か月前議論

I agree. silent agent doing agentic things async is what would be helpful, not requiring a modification to the main prompt

kimjune01

3 か月前議論

this is a patch on top of the broken flat-compaction caching algorithm used by coding agents. Why not fix the cache algorithm directly? Union-find is a better impl june.kim/union-find-compaction

kimjune01

3 か月前議論

hi Thariq, I dont know how else to reach someone at claude code, so here goes:

I solved context compaction by using a better caching algorithm. It's being implemented in gemini-cli with limited success.

https://june.kim/union-find-compaction https://github.com/google-gemini/gemini-cli/pull/24736

kimjune01

3 か月前議論

ishitani

投稿

Auditing DeepSWE

Show HN: 97% on SWE-bench Verified with subscription-token agents

56% merge rate on 316 cold OSS PRs in a week

Show HN: I submitted 316 AI-generated PRs to open source

Continuous context compaction using Union-find

Show HN: Union-find for chatbot memory instead of flat compaction

Show HN: How ads should work on chatbots, and why no one has built it yet

The Journey of Internet Ads – A Scrollytelling

Show HN: Cord – Agents that build their own coordination trees

Geometrically Allocated Ads in AI Conversations

Learn Pinyin with Games

コメント