not-chatgpt

We live in ridiculous times. Maybe being a parody is long term value creation and worth $30M from VCs.

Full thread with the receipts: https://xcancel.com/NotOnKetamine/status/2058310362647371956

Great premise but can't really agree with the execution. Felt like this makes too many implicit assumptions about LLM capabilities and traps without differentiating enough between a smart human vs AI.

Security should be a non issue in the age of AI now that auditing is cheaper than ever.

I'd give them more credits if they use the AI slop unmaintainability argument.

Been looking for a solution exactly like this but I struggle to see how it is different than spinning 10 Atlas tabs with a 2 sentence prompt.

What would be the state of the art?

This guys recent startup endorsement is the infamous rabbit r1.

Should tell you enough lol.

Cool idea. Never gonna work. LLMs are still generative models that spits out training data, incapable of highly abstract creative tasks like research.

I still remember all the GPT-2 based startup idea generators that spits out pseudo-feasible startups.

> but man this is one of the most "doomed to fail" things I've ever seen.

It will. I don't know which VC did the due diligence but the CEO once founded a scam company called RavenTech and the CTO just dropped out of CMU.

They used to be called CyberManufacture Co. and was selling NFTs. You see where this is going?

Worked at a couple FAANG companies and from what I've seen, management is absolutely right that remote work, overall, is inefficient. There was a Blind poll and I believe half the people admitted they were slacking off more frequently while remote.

However, being laid off is just a completely different matter. Depending on the company, the motivation for the layoff could significantly differ. Some did it to save costs, others have quotas to meet, etc. But I believe it is fair to say there is no observable correlation between remote work and chances of being laid off.

The zoom out is just chef's kiss

Could be that they assumed it must be a photo and the smudge was done intentionally as an artistic expression? The art world is so out of touch now I wouldn't be surprised if that was the case.

Is there an actual source for that or are you just here to complain about Stable Diffusion users flooding your Twitter with AI waifus?

If David is such a "big dreamer hippie dork," he should: 1. either open source the model or at least make it cheaper ($8/month for just 200 images is ridiculous) and 2. not bow down to dictators, or just stay out of political subjects completely.

I don't know, but China is not posing fines up to $1M and threatening 20 years in prison + civil forfeiture to just run a VPN.

Also, are you saying genocides is not one of US's main export?

In order to beat China, we need to become something worse than China?

Pretty interesting discussion - I particularly liked the part about how, in order to better align LLMs, we need better transparency so everyone can help study and align them.

A simple hack that blocks OpenAI's backend account check API helps restore current chat history.

I've tried most of these and I gotta say, ChatGLM is actually pretty good. The Chinese responses are sometimes indistinguishable from those of ChatGPT, and it's only a 6B model.

Sadly, whenever I use harder questions or logical ones, it performs terribly. It also occasionally uses Chinese adjectives (albeit correctly) when responding in English.

You still need high quality photos to generate these headshots. Anyone can just put those photos in midjourney or stable diffusion, and with ~10 minutes of prompting, generate their stylized headshots?

To generate realistic looking headshots, you need high quality training data (i.e. high quality headshots) of the person. Then what's stopping them from using stable diffusion and stylize those existing photos in the first place?

This feels like an aggressive push to force codex users to adapt to GPT3.5/4. I'd imagine they will be integrating some of codex's features into GPT4 soon.

> Someone asked a question on reddit, and you're trying to push it hard as a rumor

I am simply providing information to the best of my knowledge - it's the title of the post. The caching issue is also a guess that no one can verify yet.

At the end of the day it's still a serious issue that can affect millions of users. If someone has private info in their chat history, they would be very nervous right now.

This is due to rumored hacks/leaks: https://www.reddit.com/r/ChatGPT/comments/11wkw5z/has_chatgp...

We really need an open source alternative instead of putting all our faith in one for-profit company.

That's expected as that's how transformers are trained. I do have to clarify that it's not likely to be other people's chat session, but more likely to be training samples.

It's a poison attack that randomizes the output, but it does not allow the attack to manipulate the outputs.

For anyone looking to train on a specific style, this algorithm is useless. For any organization looking to scrape images on a massive scale, it doesn't matter as there are more than enough unaltered images out there.

It's a political solution that capitalized on fear, yet does not offer tangible protection. Even watermarking is more effective.

The description (as well as the project as a whole) is incredibly disingenuous and capitalized heavily on artist's fear of being trained by AI.

Even one pass of SD's img2img with low denoise is enough to bypass this data poisoning attack. It's a useless attack that makes training a tidbit more inconvenient.

The project also gained infamy when it stole code from an open source project with GPL license without giving credits.

Additionally, SD now offers opt outs for artists and MD likely does not train on these artists at all, and there are way more effective ways to protect art than using this algorithm.

The description (as well as the project as a whole) is incredibly disingenuous and capitalized heavily on artist's fear of being trained by AI. Even one pass of SD's img2img with low denoise is enough to bypass this data poisoning attack. It's a useless attack that makes training a tidbit more inconvenient.

The project also gained infamy when it stole code from an open source project with GPL license without giving credits.

Additionally, SD now offers opt outs for artists and MD likely does not train on these artists at all.

What does your team do? It feels like GPT4 can handle any task out there. Only drawback is latency and cost.

These articles are way more hype than substance. The review process is barely sufficient to even give credit for the AI to actually have come up with the language, not that its training data was contaminated in some way.

Better to wait for a thorough review on GPT-4's zero shot abilities from an academic paper.