Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Operator research preview

openai.com

Post reply on HN
Print view View original

141–150 of 448 posts

Re: Operator research preview

#141

Post by fsndz »

minimaxir wrote: post #8

Overall, Operator seems the same as Claude's Computer Use demo from a few months ago, including architecture requiring user to launch a VM, and a tendency to be incorrect: https://news.ycombinator.com/item?id=41914989 Notably, Claude's Computer Use implementation made few waves in the AI Agent industry since that announcement despite the hype.

This is mainly to reclaim mindshare from DeepSeek that has done incredible launches recently. R1 was particularly a strong demonstration of what cracked team of former quants can do. The demo of Operator was nice but I still feel like R1 is the big moment in the AI space so far. https://open.substack.com/pub/transitions/p/openai-launches-...

Re: Operator research preview

#142

Post by cosmic_cheese »

skydhash wrote: post #81
brap wrote: post #31

I don't know why, but the approach where "agents" accomplish things by using a mouse and keyboard and looking at pixels always seemed off to me. I understand that in theory it's more flexible, but I always imagined some sort of standard, where apps and services can expose a set of pre-approved actions on the user's behalf. And the user can add/revoke privileges from agents at any point. Kind of like OAuth scopes. Ima…

> I always imagined some sort of standard, where apps and services can expose a set of pre-approved actions on the user's behalf OS specific, but Apple has the Scripting Support API [0] and Shortcut API for their app. Works great. [0]: https://developer.apple.com/documentation/foundation/scripti...

AppleScript support has sadly become more rare over time though, as more and more companies dig motes around their castles in effort to control and/or charge for interoperability. Phoned-in cross platform ports suffer this problem too.

Re: Operator research preview

#143

Post by thrance »

transcriptase wrote: post #133

[flagged]

Can't you fathom Trump being criticized organically? You know, the convicted felon that just pardoned the neonazis that attempted to overthrow the government 4 years ago, that undid 60 years of progress by repelling the Civil Rights Act, that appointed his corrupt billionaire friends to every positions in this government but then rants on about how "DEI is taking away our meritocracy".

Re: Operator research preview

#144

Post by ActorNightly »

OoTheNigerian wrote: post #41

I'm surprised folks on Hackernews are always critical of V1s. In 18 month, apps will have APIs for "agentic browsing" ™OoTheNigerian ;) And you will not need to give anything control over your browser. I you will merely connect your app to OpenAI or any other client.

When 4o came out with its chain of thought, people thought this is it. And today, nobody really cares. Its just another LLM.

Same thing with this.

The other day I was writing some code to compute some geometric angles, and I was getting 2 different results for what I though was the same angle, but in fact I didn't realize that these angles should not be equivalent. No LLM was able to tell me the issue, they just said double check my work.

Re: Operator research preview

#145

Post by moffkalast »

easterncalculus wrote: post #22

From the slide deck on the livestream: "[Operator safety risks and mitigations] Harmful tasks: User is misaligned" Looking forward to seeing some more of the examples for when openai considers their users as "misaligned", whatever that actually even means anymore.

OAI has decided to stop aligning models and focus on aligning the users instead.

Re: Operator research preview

#146

Post by karmasimida »

fsndz wrote: post #141
minimaxir wrote: post #8

Overall, Operator seems the same as Claude's Computer Use demo from a few months ago, including architecture requiring user to launch a VM, and a tendency to be incorrect: https://news.ycombinator.com/item?id=41914989 Notably, Claude's Computer Use implementation made few waves in the AI Agent industry since that announcement despite the hype.

This is mainly to reclaim mindshare from DeepSeek that has done incredible launches recently. R1 was particularly a strong demonstration of what cracked team of former quants can do. The demo of Operator was nice but I still feel like R1 is the big moment in the AI space so far. https://open.substack.com/pub/transitions/p/openai-launches-...

R1 is a fundamental blow to their value proposition right now, the uniqueness is gone, and forever open sourced. Unless o3 is the game changer of game changer, I am not seeing they are getting the narrative back soon.

Re: Operator research preview

#147

Post by jumploops »

ActorNightly wrote: post #134
jumploops wrote: post #128

Curious how long this paradigm (computers using human interfaces) will last for P95 tasks. If the machines are smart enough, shouldn’t they be able to build better interfaces to existing software? With that aside, it seems like there are two things at play in this demo: 1. Pixel-tuned GPT-4o 2. “Agent” in prod (supervisor loop + operator loop) Will be interesting to see if they open those up as separate tools in the…

>If the machines are smart enough, shouldn’t they be able to build better interfaces to existing software? There is no "intelligence" in any of this. Just a whole lot of automation.

I used GPT-4 (entirely) to convert a Vimium-based browser control project from Python to Typescript[0].

Unlike this demo, it uses a simpler interface (Vim bindings over the browser) to make control flow easier without a fine-tuned model (e.g. type “s” instead of click X,Y coords)

I was surprised how well it worked — it even passed the captcha on Amazon!

[0] https://github.com/jumploops/vimGPT.js

Re: Operator research preview

#148

Post by cubefox »

timabdulla wrote: post #126
cubefox wrote: post #122

Earlier quoted context omitted.

> OpenAI is merely matching SOTA in browser tasks as compared to existing browser-use agents. No. It's not matching them, it's clearly exceeding them. The previous post provided the numbers.

Those numbers are not the full story. Note that GP specifically says: "Big jumps in benchmarks from _Claude's Computer Use_ though." Claude Computer Use was not SOTA for browser tasks at the time of its release (and is still not.) In WebArena, Operator does 58.1%. Previous SOTA for browser-use agents is 57.1%. In WebVoyager, Operator does 87.0%. Previous SOTA for browser-use agents is the exact same. See here for det…

Those two were two different models (Kura and jace.ai), and one model being SOTA at one benchmark doesn't make it SOTA overall. Moreover, both are specific for browser use, so they don't operate only on raw pixels but can read HTML/DOM, unlike general computer use models which rely on raw screenshots only.

Re: Operator research preview

#150

Post by _jayhack_ »

janwilmake wrote: post #121

I strongly believe we need to use Open APIs for agents. OpenAPI is the perfect specification standard that would allow for an open world and an open internet for agents. When OpenAI first came out with their first version of GPTs, it was all based on open APIs. Now they are moving away from it more and more. This means they want to control the market because they don't want to base it on an open standard. It's such a…

Unfortunately a lot of the things we want agents to interact with don't expose neat APIs. Computer use and, eventually, physical locomotion are necessary for unlocking agent interactivity with the real world.
Post reply on HN