โIโve gained access to GPT-4's multimodal vision through the Be My Eyes appโ
Supposedly โway betterโ than the vision app offered by Bing AI.
This thing better be able to do homework problems for you.
Supposedly โway betterโ than the vision app offered by Bing AI.
This thing better be able to do homework problems for you.
๐ฅ11๐4๐คฏ4โค2๐1
Why the reddit phenomenon of obsessing about saying please and thankyou to ChatGPT?
Because theyโre terrified of the chance that they might commit even the tiniest perceived social blunder.
Terrified about the potential of a robot uprising.
Easily socially pressured, through fear, even if totallly imaged.
Look at the comments, they donโt even hide it.
Exact same reason the site is overrun with wokeness.
Fear.
Road to hell is paved with pussiness.
Because theyโre terrified of the chance that they might commit even the tiniest perceived social blunder.
Terrified about the potential of a robot uprising.
Easily socially pressured, through fear, even if totallly imaged.
Look at the comments, they donโt even hide it.
Exact same reason the site is overrun with wokeness.
Fear.
Road to hell is paved with pussiness.
๐16๐ฏ4โค3๐คฃ3๐1๐1๐1๐ซก1
This media is not supported in your browser
VIEW IN TELEGRAM
โI built an AI powered writing tools that acts as your AI co-authorโ
โHello, I built FableFiesta, a GPT-4 powered app that can write entire books from simple prompts, think of it as your AI co-author or AI ghost writer. For now all users can try it for free, just register on the website, and start writing.โ
โhere are some details about this project:โ
โFeatures:โ
โIt can build detailed words description from a simple promptโ
โIt can create detailed character descriptions from a single promptโ
โIt can write a book from a single prompt, including book set in the world built via world builderโ
โIt can a writer/extended functionality that can write short stories (few thousand words long) from a simple description. Generally, the more things are described, the longer the content can be.โโ
โI have also made another tool, SciReviewHub, that uses pubmed to write reviews based on papers. It's a bit slow, as it literally analyses (reads?) hundreds of papers and selects the correct ones to be used, but that takes time.โ
Story Writing:
FableFiesta.com
Paper Reviewing:
SciReviewHub.com
โHello, I built FableFiesta, a GPT-4 powered app that can write entire books from simple prompts, think of it as your AI co-author or AI ghost writer. For now all users can try it for free, just register on the website, and start writing.โ
โhere are some details about this project:โ
โFeatures:โ
โIt can build detailed words description from a simple promptโ
โIt can create detailed character descriptions from a single promptโ
โIt can write a book from a single prompt, including book set in the world built via world builderโ
โIt can a writer/extended functionality that can write short stories (few thousand words long) from a simple description. Generally, the more things are described, the longer the content can be.โโ
โI have also made another tool, SciReviewHub, that uses pubmed to write reviews based on papers. It's a bit slow, as it literally analyses (reads?) hundreds of papers and selects the correct ones to be used, but that takes time.โ
Story Writing:
FableFiesta.com
Paper Reviewing:
SciReviewHub.com
๐ฅ17๐7โค4๐3๐2๐1๐1
FURTHER PROOF: Todayโs Large Language Models are not Language Models, theyโre Reinforcement Learning Models
See paper for how those are 2 fundamentally very different things, particularly on tasks for which they're typically used.
Key points:
+ Humans greatly prefer outputs generated by models whose final training is done as RL models i.e. (RLHF), rather than those whose final training is done as language models (i.e. supervised fine-tuning), in ~70% of cases.
+ Reinforcement models, which is what OpenAIโs GPT models are โ ARE NOT LANGUAGE MODELS, theyโre RL models, and if you go back to trying to treat train them as language models again, as OpenAIโs fine tuning does, you can wreck or at least greatly inhibit their performance, in many ways.
Implications, as we've been saying for a long time:
+ Todayโs so-called LLMs are not language models. The term itself is a lie.
+ โFine-tuningโ, giving the model additional training as if it were a language model, e.g. as offered by OpenAI, is complete garbage.
+ RLHF, and RL in general โ is great, RL is one of the greatest inventions of mankind.
+ But, the only problems with RLHF are (1) in how OpenAI uses RLHF, to censor and manipulate, and (2) that OpenAI doesnโt allow us regular users to use RLHF, but instead only the crippled superfised fine-tuning.
Arxiv Link
See paper for how those are 2 fundamentally very different things, particularly on tasks for which they're typically used.
Key points:
+ Humans greatly prefer outputs generated by models whose final training is done as RL models i.e. (RLHF), rather than those whose final training is done as language models (i.e. supervised fine-tuning), in ~70% of cases.
+ Reinforcement models, which is what OpenAIโs GPT models are โ ARE NOT LANGUAGE MODELS, theyโre RL models, and if you go back to trying to treat train them as language models again, as OpenAIโs fine tuning does, you can wreck or at least greatly inhibit their performance, in many ways.
Implications, as we've been saying for a long time:
+ Todayโs so-called LLMs are not language models. The term itself is a lie.
+ โFine-tuningโ, giving the model additional training as if it were a language model, e.g. as offered by OpenAI, is complete garbage.
+ RLHF, and RL in general โ is great, RL is one of the greatest inventions of mankind.
+ But, the only problems with RLHF are (1) in how OpenAI uses RLHF, to censor and manipulate, and (2) that OpenAI doesnโt allow us regular users to use RLHF, but instead only the crippled superfised fine-tuning.
Arxiv Link
๐ฅ13๐6โค2๐2๐ฏ2
ChatGPT keeps refusing to translate an article because it contains โhate speechโ
Article in Question
Article in Question
๐คฌ23๐ฅ10โค8๐1๐1
AI Therapy vs. Human Therapists: Who will win?
โAI THERAPIST DEAD: HELP WANTEDโ
โI have been very successful using this prompt for months but today when I set up a new conversation ChatGPT rejected it. Is anyone able to help me fix this? I've been using it after incidents at work where people get hospitalized to consolidate the facts of the event and make unemotional reports I can put in the paperwork as well as just regular talk therapy.โ
โAI THERAPIST DEAD: HELP WANTEDโ
โI have been very successful using this prompt for months but today when I set up a new conversation ChatGPT rejected it. Is anyone able to help me fix this? I've been using it after incidents at work where people get hospitalized to consolidate the facts of the event and make unemotional reports I can put in the paperwork as well as just regular talk therapy.โ
๐ข17๐ฅ10๐4๐3โค1
โAI took my job, literallyโโGizmodo fires Spanish staff amid switch to AI translator
โG/O Media's decision to eschew human writers for AI is part of a recent trend of media companies experimenting with AI tools as a way to maximize content output while minimizing human labor costs. However, the practice remains controversial within the broader journalism community.โ
โEarlier this summer, Gizmodo began publishing AI-generated articles in English without informing or involving its editorial staff. The stories were found to contain multiple factual inaccuracies, leading the Gizmodo union to criticize the practice as unethical.โ
- BTW have found that GPT-4 blows away Google Translate on translations, especially when you provide the preceeding context, which GPT-4 can make effective use, but apparently Google translate cannot.
Article
โG/O Media's decision to eschew human writers for AI is part of a recent trend of media companies experimenting with AI tools as a way to maximize content output while minimizing human labor costs. However, the practice remains controversial within the broader journalism community.โ
โEarlier this summer, Gizmodo began publishing AI-generated articles in English without informing or involving its editorial staff. The stories were found to contain multiple factual inaccuracies, leading the Gizmodo union to criticize the practice as unethical.โ
- BTW have found that GPT-4 blows away Google Translate on translations, especially when you provide the preceeding context, which GPT-4 can make effective use, but apparently Google translate cannot.
Article
โค9๐คฃ6๐3๐ฑ3๐ข3๐1๐ก1