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

Claude 3 model family

anthropic.com

Post reply on HN
Print view View original

141–150 of 723 posts

Re: Claude 3 model family

#141

Post by brookst »

dangond wrote: post #127
kylebenzle wrote: post #105

Earlier quoted context omitted.

1. It's an advertisement/press release, not so much an "article". 2. This would NOT be called even "AI" but "machine learning" 10 years ago. We started using AI as a marketing term for ML about a year ago.

This absolutely would be called AI 10 years ago. Yes, it's a machine learning task, but a computer program you can speak with would certainly qualify as AI to anyone 10 years ago, if not several decades prior as well.

Agree. ML is the implementation, AI is the customer benefit.

Re: Claude 3 model family

#142

Post by ethbr1 »

SubiculumCode wrote: post #75
ethbr1 wrote: post #41

Earlier quoted context omitted.

On the other hand, programmers are very expensive. At some level of accuracy and consistency (human order-of-magnitude?), the pricing of the service should start approaching the pricing of the human alternative. And first glance at numbers, LLMs are still way underpriced relative to humans.

NVidia's execs think so. It would be an ironic thing that it was open source that killed the programmer; as how would they train it otherwise? As a scientist, should I continue to support open access journals, just so I can be trained away? Slightly tongue in check, but not really.

I have a suspicion that greenfield science will be the last thing automated, at least the non-brute-force kind. AI assistants to do the drugery (smart search agents), but not pick the directions to proceed in.

Too little relevant training data in niche, state of the art topics.

But to the broader point, isn't this progress in a nutshell?

(1) Figure out a thing can be done, (2) figure out how to manufacture with humans, (3) maximize productivity of human effort, (4) automate select portions of the optimized and standardized process, (5) find the last 5% isn't worth automating, because it's too branchy.

From that perspective, software development isn't proceeding differently than any other field historically, with the added benefit that all its inputs and outputs are inherently digital.

Re: Claude 3 model family

#143

Post by appplication »

HarHarVeryFunny wrote: post #61
Workaccount2 wrote: post #20

Surpassing GPT4 is huge for any model, very impressive to pull off. But then again...GPT4 is a year old and OpenAI has not yet revealed their next-gen model.

Sure, OpenAI's next model would be expected to regain the lead, just due to their head start, but this level of catch-up from Anthropic is extremely impressive. Bear in mind that GPT-3 was published ("Language Models are Few-Shot Learners") in 2020, and Anthropic were only founded after that in 2021. So, with OpenAI having three generations under their belt, Anthropic came from nothing (at least in terms of models -…

What this really says to me is the indefensibility of any current advances. There’s really cool stuff going on right now, but anyone can do it. Not to say anyone can push the limits of research, but once the cat’s out of the bag, anyone with a few $B and dozen engineers can replicate a model that’s indistinguishably good from best in class to most users.

Re: Claude 3 model family

#144

Post by numeri »

michaelt wrote: post #112
nine_k wrote: post #100

Earlier quoted context omitted.

I personally find approaches like this the correct way forward. An input analyzer that finds out what kinds of tokens the query contains. A bunch of specialized models which handle each type well: image analysis, OCR, math and formal logic, data lookup,sentiment analysis, etc. Then some synthesis steps that produce a coherent answer in the right format.

Then you might enjoy looking up the "Mixture of Experts" model design.

That has nothing to do with the idea of ensembling multiple specialized/single-purpose models. Mixture of Experts is an method of splitting the feed-forwards in a model such that only a (hopefully) relevant subset of parameters is run for each token.

The model learns how to split them on its own, and usually splits based not on topic or domain, but on grammatical function or category of symbol (e.g., punctuation, counting words, conjunctions, proper nouns, etc.).

Re: Claude 3 model family

#145

Post by brookst »

spyder wrote: post #99

What's up with the weird list of the supported countries? It isn't available in most European countries (except for Ukraine and UK) but on the other hand lot of African counties are listed... https://www.anthropic.com/claude-ai-locations

EU has chosen to be late to tech in favor of regulations that seek to make a more fair market. Releasing in the EU is hard.

Re: Claude 3 model family

#147

Post by SubiculumCode »

ethbr1 wrote: post #142
SubiculumCode wrote: post #75

Earlier quoted context omitted.

NVidia's execs think so. It would be an ironic thing that it was open source that killed the programmer; as how would they train it otherwise? As a scientist, should I continue to support open access journals, just so I can be trained away? Slightly tongue in check, but not really.

I have a suspicion that greenfield science will be the last thing automated, at least the non-brute-force kind. AI assistants to do the drugery (smart search agents), but not pick the directions to proceed in. Too little relevant training data in niche, state of the art topics. But to the broader point, isn't this progress in a nutshell? (1) Figure out a thing can be done, (2) figure out how to manufacture with human…

I think that picking a direction is not that hard, and I don't know that AI couldn't do it better. I'm not sure mid-tier CEO's won't be on their way out, just like middle management.

Re: Claude 3 model family

#148

Post by glenstein »

Alifatisk wrote: post #25
sidcool wrote: post #6

Wow. 1 million token length.

Yeah this is huge, first Gemini and now Claude!

Right, and it's seems very doable. We've been getting little bells and whistles like "custom instructions" have felt like marginal addons. Meanwhile huge context windows seem like they are a perfect overlap of (1) achievable in present day and (2) substantial value add.

Re: Claude 3 model family

#149

Post by staticman2 »

avereveard wrote: post #35
_sword wrote: post #14

At this point I wonder how much of the GPT-4 advantage has been OpenAI's pre-training data advantage vs. fundamental advancements in theory or engineering. Has OpenAI mastered deep nuances others are missing? Or is their data set large enough that most test-cases are already a sub-set of their pre-training data?

So far gpt is the only one able to answer to variations of these prompts https://www.lesswrong.com/posts/EHbJ69JDs4suovpLw/testing-pa... it might be trained on these but still you can create variations and get decent responses Most other model fail on basic stuff like the python creator on stack overflow question, they identify Guido as the python creator, so the knowledge is there, but they don't make the connection…

>>So far gpt is the only one able to answer to variations of these prompts

You're saying that when Mistral Large launched last week you tested it on (among other things) explaining jokes?

Re: Claude 3 model family

#150

Post by causal »

virgildotcodes wrote: post #68

Just signed up for Claude Pro to try out the Opus model. Decided to throw a complex query at it, combining an image with an involved question about SDXL fine tuning and asking it to do some math comparing the cost of using an RTX 6000 Ada vs an H100. It made a lot of mistakes. I provided it with a screenshot of Runpod's pricing for their GPUs, and it misread the pricing on an RTX 6000 ADA as $0.114 instead of $1.14.…

I'm guessing the difference is screenshot reading, I'm finding that it's about the same as GPT-4 with text. For example, given this equation:

(64−30)−(46−38)+(11+96)+(30+21)+(93+55)−(22×71)/(55/16)+(69/37)+(74+70)−(40/29)

Calculator: 22.08555452004

GPT-4 (without Python): 22.3038

Claude 3 Opus: 22.0492

Post reply on HN