Cohere: The world’s most powerful NLP toolkit

After clicking through endless marketing speak on multiple pages I still wasn't able to find anything remotely substantiating the claim "the world's most powerful'.

As it is running on cloud machines as a SaaS business that might theoretically just mean potentially most available computer cores for a comparable offering.

Or any other BS market g can come up with. At the time of writing there isn't even a footnote or anything to that claim. Companies that do this are on my instant black list. Why should I trust them, when they don't show sources, or anything to back up their claim in an easy to discover way.

Edit before potential answers: Maybe they have something to the claim. Maybe the offering is actually great. I wouldn't know. I wasn't able to somewhat verify the big claim and therefore did not go deeper into the offering. Unsubstantiated marketing speak just rubs me the wrong way.

I just want to note the replies to this thread are excessively dismissive and toxic. You may not agree with the wording of their advertising ("world's most powerful NLP toolkit" is marketing speak, sure) but going from that to implying the technical side is "only Min-GPT" is tremendously weird. As someone who works in machine learning and specifically language models this is a team I'm keeping an eye on.

For anyone who wanted more technical discussion re: ML / LM (though the author notes this work "[does] not reflect the architectures or latencies of my employer's models" i.e. it's an exploratory technical breakdown of general model characteristics) I've appreciated the technical write-ups from @kipperrii (ML ops @ Cohere) recently:

- Transformer Inference Arithmetic: https://carolchen.me/blog/transformer-inference-arithmetic/

- Breakdown of H100s for Transformer Inferencing: https://carolchen.me/blog/h100-inferencing/

I think this post was a trick question that you have in exams. The sentiment in this thread is negative (at the point in time I am posting). You can say whatever you want, they ended up raising a lot of capital from well known VC firms [1] and has few of the smartest guys working there. Does that mean the org is legit, I don't know. Does that mean their claims are valid, I don't know. Jumping into a negative conclusion without doing a proper due diligence is something I am seeing lately.

Criticism is different from downright pessimism. I know founders and entrepreneur should have thick skin, but few people who could be potential customers or investors take these threads and the sentiments in hacker news as grain of salt.

[1] https://www.crunchbase.com/organization/cohere-82b8

Hey all, I'm Aidan! I cofounded Cohere.

Really appreciate the folks hyping up our team heheh

Also the feedback on the marketing copy being shit really resonates. We're going to be launch a refresh pretty soon which I'll make sure reads better.

It would be super helpful to me if some of you here would be willing to take an advanced look and just tear it up with some critical feedback!

Also anytime you have feedback about the product; like model quality, the experience of onboarding and using the product; etc. please shoot me an email!!

I'm at aidan.gomez@cohere.ai

Thanks again all!

BS marketing aside, there is a real market for Cohere, GPT-3 and other large language model APIs. Tonnes of companies want to generate embeddings from text or perform other NLP tasks, without having to hire machine learning experts. Why handle the infrastructure, training data, and model training yourself? Let someone else handle that. POST your text and get an embedding vector back.

NLP is going the same way as the rest of "cloud", towards managed services. Plenty of companies will roll their own ML models. If NLP is a core part of your business it might make sense to hire a machine learning team. If NLP is just one small feature that might be scrapped in 3 months after an A/B test shows negative results, using a managed service is a no brainer.

I went to university with these folks, they worked with Hinton and wrote the transformers paper (T in GPT) - just wanted to mention that there is real tech behind all of the marketing speak

World most powerful data scientist here. I disagree.

The problem with Cohere is that it’s an “all-in” platform. They want to control everything from data collection to training to ops. It’s the most lock-in thing you can do from a data science perspective and intends to compete with Sagemaker.

OK, I just don't trust "the world's most".

I wonder, why are you calling yourself the most powerful ?

Any comparison to solutions from Hugging Face or John Snow Labs ?

I know these guys have been claiming SOTA for a few years and don't see how Cohere would be any more powerful, do you provide training ?

This is not to be mistaken with the company https://cohere.io/ that lets you monitor your web users in real-time.

For the discussion forums spam detection use case, any service recommendation (or a open lib model) to detect if a user submitted comment/post is spam/advertisement

This is probably an API gateway fronting Spark NLP and a mini GPT. They have 25 engineers, what do they all do?

This is a SaaS.

How modest of them. As of lately, state of the art lasts only one week in NLP.

Well, Jay Alaamar (one of the greats of our field) is working there so I automatically believe in them. Too bad Jay won't connect to me on LinkedIn, and likely played a part in my resume being rejected before the initial phone screen!

My own heros think I am nothing!