gvkhna

Pi is without contest the best harness. Because it’s yours, the total control is the only way to move from driving agents to orchestrating them.

My interruptions and frustration with llms has gone down a cliff. It’s amazing to see what just a few lines of system prompt really can do to keep agents doing the correct thing much more often.

The harness is making a big difference, lackluster performance with pi but somehow very good performance with opencode. There’s some rl there for sure, for a smaller model it’s likely going to perform much better in a harness it understands the best.

256gb ddr5, so this is going to cost north of 100k? Damn.

While possible the amount of variation in serving infrastructure is unlikely to land with actually giving the exact same errors zhipu does.

It feels like glm flash, and there was a report zhipu had secured a huge new cluster suggesting they have the capacity. My guess anyway.

https://www.tomshardware.com/tech-industry/artificial-intell...

If it’s not zhipu then why is it returning errors that zhipu does for other models? Who else would return the exact same errors even if they took a lot of core infra like tokenizer from z?

That’s the whole point, just cost and compute limitations in your way (mostly).

Incidentally the fact that you could read and write from essentially thousands of harddrives at once for even a tiny file makes you wonder what novel use cases you could get out of that.

Nice @OP i put together something similar as well. Incidentally I found for motion design specifically llm is not able to infer specific animations as well as it just being described very plainly and accurately what is happening and the timing.

One thing which sort of worked decently was actually take the frames and put them into a grid and have the agent look at the image of all of the frames together. It did surprisingly well but missed a lot of subtle details that it couldn’t see.

Also tried various kinds of vision embeddings, heat map of motion etc, and blur etc to show motion. But none really worked as well so I ended up just describing it until it got it. Haven’t quite found the right solution yet.

Go is my go to for simple coding projects now, if i have an agent work grind on it. It’s lightweight fast and resource minimal, while being quick to code.

Go tooling is kind of surprisingly minimal and takes more taming then should be required. Especially monorepo workspace support seems a bit like an afterthought.

Hope to see continued improvements because in agentic era go is a serious go to.

Rust is top tier but for most things, it’s not really needed for an agent, to spend longer tokens fixing memory ceremony.

But rust tooling is best in class overall I feel and hope google continues improvements for go.

Use the real browser: clawchrome.com. You can evaluate code on the page, do network capture, screenshot. But it’s the real browser, no fork, no cdp.

For lower RTT you’ll need an instance and proxy close to upstream but a real browser will help with scraping detection.

UUIDv7 as pk is inherently the best option besides just a bigint where you really don’t need a public id.

But a Url62 as a url safe public id from the pk is simple and straightforward to use and comes with few risks of leak issues. Wish postgres had native base62 encoding for url62 now that it has uuidv7 native.

I agree technically but in most use cases the timestamp from uuidv7 is not a security leak. Especially where you’re already sharing that data in some way or another. A default guid is unnecessary if you use uuidv7 I think (in most situations).

Clawchrome uses a mitm proxy like this for network capture of a browser. I had my agent test using network capture with playwright, then again with simple HAR style reconstruction of the mitm proxy capture. It was just as effective.

Check it out if you want to give your agent a real browser, no cdp. clawchrome.com

That’s at the request of a user they aren’t operating autonomously.

There are many usecases for agents having access to a real browser and many usecases for humans where work wise it's useful or for agent success in completing the task. Including any specific situations where sites may block "controlled" browsers, as well as co-working.

If an agent is controlling your own browser, it can get in the way of you being able to do work as well.

Additional security control, if you have control and audit trails regarding what the agent is doing, think things like not having access to the keys, cookies are unavailable through the api, certain sites blocked, especially for agents that run automated background tasks.

There's a ton of reasons really. Using stealth chromium forks can decrease security, potentially get you blocked from services/sites, and typically can be detected easily because Google is doing a lot more than just compiling Chromium to release Google Chrome now. That's why many of the stealth browsers out there are moving targets, and have tons of instances of being blocked.

You can trace almost anything back to incentives, renewables are more profitable now in one way or another.

Interesting that no one is talking about identity verification likely coming anyway. I’m working on clawchrome.com, a real browser for agents. It can access any website because it’s the real browser.

I sure hope agents don’t swarm social media. But at the same time I think identity verification companies have a tougher problem, ai can produce real looking videos and documents. There’s probably no real way to verify someone purely through the internet at this point.

Unfortunately AT-SPI support is very hit or miss, mostly miss. To get it useable for this required a huge amount of modification. See my other comment for context.

Built this but not open source because it’s more designed for the bot evasion/stealth market. But it’s designed to let ai control the real browser without cdp. So no chromium forks/nothing to “detect.”

In this case with the, I believe it’s called quantum tunneling by mullvad, it’s actually a good thing. Because the encapsulation protocol is just UDP/IP, a well established existing protocol that can masquerade as any kind of internet traffic easily.

Makes it difficult to block by censors. Great video I saw here: https://youtu.be/pZiG8r-diTM?si=wy35elqMt1T6euq0

This also means wg is just doing one thing instead of a dozen it doesn’t “need” to.

No I haven’t but I think the lack of liquidity as a chicken and egg is a huge barrier to entry in these markets specifically. They are small right now but there are climate derivatives on the Chicago mercantile exchange so this isn’t a new concept I think.

Could you tell me more? https://discord.gg/HPpN42SKQ

Working on AI driven weather predictions to make money on prediction markets. The accuracy of WeatherNext 2 is astounding.

It may be a fools errand but makes for an extremely interesting research project. http://climatesight.app if you’re interested in climate markets.

I think it’s going to be targeting mostly stationary HA redundant uplinks. Backup for primary uplink or low usage primary link. In those scenarios pointing at your nearest antenna fixed is much better than an omnidirectional antenna.

Working on this problem but a combo of a skill and an mcp better suited to playwright is the solution IMO.

The issue is for many things playwright is really verbose, by better tailoring outputs and making them more fine grained you’ll get less context bloat and allow the llm to better work with the context. I’m making it open source.

Nice my project also uses Deno for sandbox and bun runtime because it just works.

http://github.com/gvkhna/vibescraper

I told Jared if bun had a sandbox i’m pretty sure everyone would use it.

Ah interesting, Yes i see what you’re saying regarding eslint and didn’t consider that. I’m a big of newer linters with better perf but none match the depth currently with eslint. Will keep an eye out on deno lsp!

Yup, github.com/gvkhna/vibescraper in the sandbox package. Def needs a cleanup but it’s heavily tested and works great at aivibescraper.com

Github actions buildx also going down is a really unintended consequence. It would be great if we could mirror away from docker entirely at this point but I digress.

Wow, this is the key. If it just had python that’s not as useful but the major frameworks is the real value. Definitely going to keep an eye on this. I built a sandbox with deno for ai code generation. It works well enough but there are some use cases where python may make more sense. Nice!

It was sophisticated by using github notifications.

But shitty because it said W25 is a 100m cohort.