Hacker News Reader

Comment by wodenokoto | Hacker News Reader

I’ve never heard of anthropic. Who are they and what do they offer that’s worth that investment?

They have an excellent LLM called Claude which is quite competitive with OpenAI offerings and has long context windows.

Other than OpenAI’s models they have been able to deliver the best results in projects I worked on and sometimes they outperformed OpenAI models too.

I’m happy to see this as this gives them funding to further train expensive models and compete - their product is really nice at the moment but could really benefit from bigger size and even bigger contexts.

They have Claude 2 which is up there with GPT-4 and PaLM 2 as one of the top three publicly accessible large language models. And presumably those billions will train Claude 3.

I'm sorry but I'll have to disagree with you - Claude 2 is worse than GPT-4 in almost all tasks I've tried it with, the two things that it's better at than GPT-4 are the context window and the dataset being from early 2023 instead of late 2021 (for 3.5 Turbo/4).

Being worse than GPT-4 doesn't dispute GP's claim that it's in the top 3 with GPT-4. To disagree with them you need to name models other than GPT-4 and PaLM 2 which are also better than Claude 2.

I've found it to be marginally worse at some coding tasks than gpt-3.5, but it is all highly specific to the kinds of questions that I'm asking.

They are basically the only direct competitor to OA.

Except for Google

Google Deepmind exists. All 3 are trying to build a generalist AI agent, as opposed to the more specialised models of say CharacterAI or inflection.

Is Deepmind doing LLMs? I thought it was a different team in Google that trained Bard/Palm

They merged Deepmind into Brain and rebranded it all Deepmind.

I think they merged the teams

I’m still trying to get my mind around a world where Brain/DeepMind aren’t way ahead on all this. They were so far ahead so recently.

Is there a comeback story here? IIRC they’ve still got Dean and Hassabis and a bunch of other bruisers, along with arbitrary financing, the best training sets, and novel compute.

Is it time to put Hassabis in charge and start making Ilya and Karpathy offers they can’t refuse?

The world's greatest engineering talent won't save you if the management is incompetent.

Google may yet experience a comeback, but it will start with throwing out the current C-suite.

Hassabis has some pretty serious chops on the leadership side as well as a firm grip on the technology.

Him as CEO?

Deepmind didn't produce any AI product with good proven adaptation, while burned many billions.

Sure, they're all competing labs, but Deepmind isn't a competitor for a product offering. Anthropic is quite analogous to OpenAI because they have an API you can pay for an a chatbot product you can use.

Reply on news.ycombinator.com