Introducing computer use, a new Claude 3.5 Sonnet, and Claude 3.5 Haiku

Introducing computer use, a new Claude 3.5 Sonnet, and Claude 3.5 Haikuanthropic.com
Introducing computer use, a new Claude 3.5 Sonnet, and Claude 3.5 Haiku
by timdaub.eth12379 🥝 • 2y
AI summary of the linked article

Anthropic announced an upgraded Claude 3.5 Sonnet, a new model called Claude 3.5 Haiku, and a public beta of "computer use," which lets developers direct Claude to operate computers by viewing screens, moving a cursor, clicking and typing. The upgraded Sonnet raised its SWE-bench Verified score from 33.4% to 49.0% and its TAU-bench retail score from 62.6% to 69.2% at the same price and speed as its predecessor. Claude 3.5 Haiku scores 40.6% on SWE-bench Verified, and the company describes computer use as experimental, still error-prone, and in early stages.

Recommended by 2 curators
avatar
Check out this demo video of their computer use feature https://youtu.be/jqx18KgIzAE?si=0zP6MiY_z5KvMQIw


pretty wild
avatar
Using Claude 3.5 Sonnet quite a lot lately. It is great to chat and order your thoughts. Also I appreciate the overall tone of voice of the output.
avatar
Is it much better that ChatGPT? I haven‘t even tried it yet, still on Chatgpt4
avatar
It depends on the use case. But ChatGPT 4o feels faster. I haven’t chatted with Claude since today’s update. It might have improved. Claude is great for in-depth philosophy, on that front it might be better. And they say Claude is better for coding. For quick info on stuff ChatGPT 4o might be better. Yeah they’re both amazing chatbots.
avatar
Anthropic released have been so wild recently. I can actually think of quite a few use cases of Computer use, but ideally those would happen in a VM that I run in my existing OS so that the AI doesn‘t gain access to my data or credentials and accidentially fucks up.

Although, the Googling use case is kinda stupid. Asking an LLM directly is a better use of the tokens
avatar
Lately have been trying out Kagi Assistant which is private by default. Besides Anthropic models, they also have OpenAI, Mistral, Gemini and Llama. And you can mix the models and create your own custom assistant. I like it. (And you can pay with Bitcoin.)


Since we don’t show ads (and never will) and don’t train on subscriber data there’s no reason for us to harvest your data, track your clicks, searches, threads, or build a profile of you. When we use third party models via their APIs it is protected under terms of service that forbid using data for training their models (e.g. Anthropic Terms & Google Terms).
Characters remaining: 10,000

comment guidelines
HomeNewSubmit