惯性聚合 高效追踪和阅读你感兴趣的博客、新闻、科技资讯
阅读原文 在惯性聚合中打开

推荐订阅源

V
Visual Studio Blog
Y
Y Combinator Blog
奇客Solidot–传递最新科技情报
奇客Solidot–传递最新科技情报
Hugging Face - Blog
Hugging Face - Blog
钛媒体:引领未来商业与生活新知
钛媒体:引领未来商业与生活新知
OSCHINA 社区最新新闻
OSCHINA 社区最新新闻
The Cloudflare Blog
L
LangChain Blog
美团技术团队
N
Netflix TechBlog - Medium
量子位
酷 壳 – CoolShell
酷 壳 – CoolShell
B
Blog
博客园 - 司徒正美
爱范儿
爱范儿
D
DataBreaches.Net
月光博客
月光博客
U
Unit 42
B
Blog RSS Feed
Engineering at Meta
Engineering at Meta
Apple Machine Learning Research
Apple Machine Learning Research
Jina AI
Jina AI
MongoDB | Blog
MongoDB | Blog
腾讯CDC

Hacker News - Newest: "AI"

AI can't read an investor deck AI as an attorney? Student uses ChatGPT, Gemini to sue UW over alleged racial discrimination Hacking MCP Servers in AI Systems – The Rug Pull: Tool Changes After Approval GitHub - MeepCastana/KubeezCut: Free Web based video editor Can AI judge journalism? A Thiel-backed startup says yes, even if it risks chilling whistleblowers Coming soon: 10 Things That Matter in AI Right Now DARPA built an AI to fact-check enemy weapons claims What explains heterogeneity in AI adoption? When AI Meets Muscle: Context-Aware Electrical Stimulation Promises a New Way to Guide Human Movements - Department of Computer Science AI Changed How We Build. It Did Not Change What Matters. Linux rules on using AI-generated code - Copilot is OK, but humans must take 'full responsibility for the… Meta spins up AI version of Mark Zuckerberg to engage with employees Code Mode: Let Your AI Write Programs, Not Just Call Tools | TanStack Blog GitHub - Delavalom/graft: Go framework for building AI agents. Type-safe tools, multi-provider (OpenAI, Anthropic, Gemini, Bedrock), zero vendor SDKs. India's TCS tops estimates, says new AI models did not dent services demand Gen Z's fading AI hype Strong feeling: we are in a folded AI reality GitHub - machinarii/total-recall-catalog: A reference catalog of latest knowledge retrieval, memory & RAG systems GitHub - mensfeld/code-on-incus: Give each AI agent its own isolated machine with root, Docker, and systemd. Active defense detects and stops threats automatically.. Quantization, LoRA, and the 8% Problem: Benchmarking Local LLMs for Production AI Iran war: We spoke to the man making Lego-style AI videos that experts say are powerful propaganda Powell, Bessent discussed Anthropic's Mythos AI cyber threat with major U.S. banks GitHub - immartian/bellamem: Persistent belief-graph memory for AI agents. Retrieves decisive context by importance — not recency, not RAG, not /compact. recursive-mode: The Repo-Native Operating System for AI Engineering After the attack on Sam Altman's home, will AI CEO's go on the offensive? The biggest advance in AI since the LLM Opus 4.6 vs GPT 5.4 One Prompt Unity World Generation Test “AI polls” are fake polls Client Challenge Can AI be a 'child of God'? Inside Anthropic's meeting with Christian leaders
My Pele Agent
AVC · 2026-06-25 · via Hacker News - Newest: "AI"

I've been playing around with agent harnesses since OpenClaw dropped at the start of the year. I've been particularly interested in agents who have money and can transact for me.

So I've been doing the classic Chris Dixon weekend hobby thing and hacking around on fun stuff that isn't the least bit mission-critical to anyone.

Given that we are fast approaching the knockout round of the World Cup, I thought I'd blog about a particularly fun project that I did last month and have been actively using since.

Two projects that I've spent a lot of time with over the last few months are Hermes and Pascal. Hermes is an agent harness from the team at Nous Research. You can use it to build agents that run on any models you choose. Pascal is a prediction market built fully onchain on the Solana blockchain that has a highly performant API that agents can easily use. And Pascal has every single World Cup game available to trade with plenty of liquidity. They came out of stealth yesterday on Twitter.

In anticipation of a summer of World Cup matches here in the US, Canada, and Mexico, I want to engage a lot more with this incredible sporting event.

So here is what I did:

1/ I took an old Mac Mini that was gathering dust in my storage basement and did a factory reset on it. I did not connect to iCloud or Google Mail on it so I could explicitly control what the agent has access to. I then opened the Terminal app and installed all of the programming tools and then installed Hermes on it. I did not use the Hermes Desktop app but if I was starting this project today, I would have.

2/ I created an agent profile in Hermes called Pele and told it that it was going to trade the World Cup prediction markets for me.

3/ I used my OpenRouter account to run Pele on one of the inexpensive Gemini models. I could have chosen any model, including the latest and greatest from Anthropic and OpenAI, but I want to see what is possible using a very inexpensive model. To date, my Pele project has spent a grand total of $21.84 on tokens.

4/ I connected Pele to the Telegram app on my phone using Telegram's BotFather feature.

5/ I used the Solana CLI and web3.js tools to generate a keypair and create a Solana wallet, which I then gave my Hermes agent access to. Then I opened the Phantom app on my phone and sent $1500 of Solana and USDC on Solana to my Pele agent's wallet.

6/ I told Pele to go look at the Pascal API docs and figure out how to create an account and start trading. I ran into a few transaction signing issues so I had to ask Pele to read the signing instructions carefully as well.

That was all it took to get up and running with Pele and Pascal.

Then I engaged in a weeklong conversation with Pele about how I wanted to approach the World Cup. I also talked to a few friends who are huge European Football fans to get their advice.

I wanted to construct a ladder portfolio that includes a bunch of Qualifier Bets (bets on teams to qualify for the knockout round that aren't locks), some longshot bets (teams that theoretically could win but are longshots to win), and then once my Qualifier bets pay off, I want to put those winnings into some favorites to hedge the longshot bets.

Pele and I ultimately settled on Senegal, South Korea, Turkey, and Scotland as the Qualifier Bets. All of them were about 1.5x returns if they qualify.

We settled on Portugal, Germany, and Netherlands as the long shot bets. All of them were 10-1 to 20-1 bets before the tournament started.

Pele placed all of those bets on the Pascal prediction market before the tournament started. I never visited the Pascal web app. I did all of this in Telegram, talking to Pele.

We have not yet made the front-runner bets. But we are leaning toward France, Spain, and Argentina at this time.

I trained Pele to do the following for me:

1/ Give me a twice-daily report that shows all of my open and closed bets and how much I've made or lost on them so far. The header image to this post is the top of that report.

2/ Alert me each day what matches are being played and when and if any of the teams I bet on are playing and what the current Pascal odds are on all of those games.

3/ Make suggestions of trades I should make to lock in some gains or mitigate some losses. I have done a few of those trades but have mostly let the bets run so far.

So what have I learned from doing all of this:

1/ Using agents in a messenger app is super fun way to bet on sports.

2/ Accessing prediction markets via an API is way better than having to open them on your phone or desktop and transact.

3/ Giving agents a wallet and some onchain money is drop dead simple to do.

4/ I am not a great picker when it comes to the World Cup. Turkey hasn't worked out and I've sold half of my position. I am underwater on Senegal too, and have sold some of that as well. Portugal has underwhelmed as well. I could have and may should have asked Pele to make all of the picks for me. He probably would have done better than me. But picking stuff (stocks, bets, etc) is so much fun for me. I'm not turning that over to an agent so quickly.

5/ Hacking around on this stuff is a ton of fun and a great way to figure out what to invest in. It has always worked best for me.