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

ChatGPT for Teams

openai.com

Post reply on HN
Print view View original

121–130 of 453 posts

Re: ChatGPT for Teams

#121

Post by novaRom »

kolinko wrote: post #101
Demiurge wrote: post #74

Earlier quoted context omitted.

I have 20 years of software development experience, and I couldn’t understand anything you said. Is there a dictionary for this new lingo, or am I just too mid?

He speaks very unclearly, instead of saying GPT-4-turbo he says 4.5 preview. 4.5 is invention of his. Also mixtral medium - no idea of what he means by that. Not to mention a claim that mixtral is as good as gpt-4. It’s on the quality of gpt3.5 at best, which is still amazing for an open source model, but a year behind openai

I think with Mixtral Medium they mean MoE 2x13B which is on top on huggingface leaderboard? It is still not close to 8x175B, but size alone is not most important factor. With smarter training methods and data it is possible we will see performance similar to gpt-4 in open source mixture of experts of smaller sizes.

Re: ChatGPT for Teams

#123

Post by nopinsight »

benreesman wrote: post #66

I’ve got my stuff rigged to hit mixtral-8x7, and dolphin locally, and 3.5-turbo, and the 4-series preview all with easy comparison in emacs and stuff, and in fairness the 4.5-preview is starting to show some edge on 8x7 that had been a toss-up even two weeks ago. I’m still on the mistral-medium waiting list. Until I realized Perplexity will give you a decent amount of Mistral Medium for free through their partnership…

Curious that you mentioned "4.5-preview". What do you mean there?

To my knowledge, and I searched to confirm, GPT-4.5 is not yet released. There were some rumors and a link to ChatGPT's answer about GPT-4.5 (could also be a hallucination) but Sam tweeted it was not true.

Re: ChatGPT for Teams

#124

Post by benreesman »

rvba wrote: post #117
benreesman wrote: post #66

I’ve got my stuff rigged to hit mixtral-8x7, and dolphin locally, and 3.5-turbo, and the 4-series preview all with easy comparison in emacs and stuff, and in fairness the 4.5-preview is starting to show some edge on 8x7 that had been a toss-up even two weeks ago. I’m still on the mistral-medium waiting list. Until I realized Perplexity will give you a decent amount of Mistral Medium for free through their partnership…

Was this generated by some AI? It it a parody?

I've made similar apologies upthread but I'm passionate about this being an inclusive conversation and so I'm trying to respond to everyone who I confused with all the jargon.

The trouble with the jargon is that it obfuscates to a high degree even by the standards of the software space, and in a field where the impact on people's daily lives is at the high end of the range, even by the standards of the software space.

HN routinely front-pages stuff where the math and CS involved is much less accessible, but for understandable reasons a somewhat tone-deaf comment like mine is disproportionately disruptive: people know this stuff matters to them either now or soon, and it's moving as quickly as anything does, and it's graduate-level material.

If you have concrete questions about what probably looks like word salad I'll do my best to clarify (without the aid of an LLM).

Re: ChatGPT for Teams

#125

Post by hanspeter »

bix6 wrote: post #20

“No training on your business data or conversations” Does this mean they will still use your data for other non-training purposes?

Yes. They will use your data as input to the GPT model to deliver the reponse you have requested.

Re: ChatGPT for Teams

#126

Post by epups »

benreesman wrote: post #84
icelancer wrote: post #71

Earlier quoted context omitted.

> Mistral Medium destroys the 4.5 preview. On what metrics? LMSys shows it does well but 4-Turbo is still leading the field by a wide margin. I am using 8x-7b internally for a lot of things and Mistral-7b fine-tunes for other specific applications. They're both excellent. But neither can touch GPT-4-turbo (preview) for wide-ranging needs or the strongest reasoning requirements. https://huggingface.co/spaces/lmsys/cha…

Keep in mind that modern quantitative approaches to LLM evaluation have been effectively co-designed with the rise of OpenAI, and folks like Ravenwolf routinely disagree with the leaderboards. There's also very little if any credible literature on what constitutes statistically significant on MMLU or whatever. There's such a massive vested interest from so many parties (the YC ecosystem is invested in Sam, MSFT is in…

> Keep in mind that modern quantitative approaches to LLM evaluation have been effectively co-designed with the rise of OpenAI, and folks like Ravenwolf routinely disagree with the leaderboards.

Sorry but you're talking complete nonsense here. The benchmark by LMSys (chatbot arena) cannot be gamed, and Ravenwolf is a random-ass poster with no scientific rigor to his benchmarks.

Re: ChatGPT for Teams

#127

Post by dudeinjapan »

benreesman wrote: post #66

I’ve got my stuff rigged to hit mixtral-8x7, and dolphin locally, and 3.5-turbo, and the 4-series preview all with easy comparison in emacs and stuff, and in fairness the 4.5-preview is starting to show some edge on 8x7 that had been a toss-up even two weeks ago. I’m still on the mistral-medium waiting list. Until I realized Perplexity will give you a decent amount of Mistral Medium for free through their partnership…

Speculative musings beckon, and we dare to embrace them. The crux of the matter appears to be the chasm that separates novel advancements from the moment they are quantified for mainstream consumption. Retaining vivid memories of past entanglements with industry titans, circumspectly explore and exploit these innovations until they become both affordable and practicable for on-premise utilization, finally unveiling competitive prowess. The overarching question looms large. Perhaps, Mistral has not yet amassed the financial resources commensurate with such largesse.

"My hips don't lie."

Re: ChatGPT for Teams

#128

Post by tikej »

benreesman wrote: post #96
Demiurge wrote: post #83

Earlier quoted context omitted.

That gave me an idea, here is what I got from Copilot: You have set up your system to run different AI models and compare their performance using a text editor. You are using Mixtral-8x7, a high-quality open-source model developed by Mistral AI, Dolphin, an emulator for Nintendo video games, 3.5-Turbo, a customized version of GPT-3.5, a powerful natural language model, and 4-Series Preview, a new version of the BMW s…

https://imgur.com/WDrqxsz

What kind of emacs distribution is on the screenshot?

Re: ChatGPT for Teams

#129

Post by benreesman »

pyinstallwoes wrote: post #73
benreesman wrote: post #66

I’ve got my stuff rigged to hit mixtral-8x7, and dolphin locally, and 3.5-turbo, and the 4-series preview all with easy comparison in emacs and stuff, and in fairness the 4.5-preview is starting to show some edge on 8x7 that had been a toss-up even two weeks ago. I’m still on the mistral-medium waiting list. Until I realized Perplexity will give you a decent amount of Mistral Medium for free through their partnership…

Can you share some examples of how you are using it? Mixtral that is? What's your setup? What's your flow/workflow?

I screenshotted my emacs session upthread in a bit of a cheeky "AI-talking-about-AI" joke: https://imgur.com/WDrqxsz.

While I heavily rely on `emacs` as my primary interface to all this stuff, I'm slowly-but-surely working on a curated and opinionated collection of bindings and tools and themes and shit for all the major hacker tools (VSCode, `nvim`, even to a degree the JetBrains ecosystem). This is all broadly part of a project I'm calling `hyper-modern` which will be MIT if I get to a release candidate at all.

I have a `gRPC` service that wraps the outstanding work by the "`ggeranov` crew" loosely patterned on the sharded model-server architectures we used at FB/IG and mercilessly exploiting the really generous free-plan offered by the `buf.build` people (seriously, check out the `buf.build` people) in an effort to give hackers the best tools in a truly modern workflow.

It's also an opportunity to surface some of the outstanding models that seem to have sunk without a trace (top of mind would be Segment Anything out of Meta and StyleTTS which obsoletes a bunch of well-funded companies) in a curated collection of hacker-oriented capabilities that aren't clumsy bullshit like co-pilot.

Right now it's a name and a few thousand lines of code too rough to publish, but if I get it to a credible state the domain is `https://hyper-modern.ai` and the code will be MIT at `https://github.com/hyper-modern-ai/`.

Re: ChatGPT for Teams

#130

Post by realusername »

I'm not too suprised by the move, it's a classic segmentation steategy but I was surprised how poorly the example screenshots they gave reflect on the product.

You have one non actionable marketing answer, a growth graph created without axis (what are people going to do with that?) and a Python file which would be easier just to run to get the error.

That kind of reinforce my belief that those AI tools aren't without their learning curves despite being in plain English.

Post reply on HN