DalasNoin

8 ms·

91.
by DalasNoin 3y ago
It is unfortunate that some people hear AI safety and think about chatbots saying mean stuff, and others think about a future system performing the machine revolution against humanity.
92.
by DalasNoin 3y ago
Why do you find this so surprising? You make it sound as if OpenAI is already outrageously safety focused. I have talked to a few people from anthropic and they seem to believe that OpenAI doesn't care at all about safety.
93.
by DalasNoin 3y ago
Is there any good evidence of the board actually wanting him back? Sure, he was invited back into the office, but maybe Mira allowed this as CEO? From the outside it seems plausible that the board was just under immense pressure but was fir
94.
by DalasNoin 3y ago
Reading the statement, I am doubtful that Microsoft and OpenAI can continue their business relationship. I think the most aggressive part of this is the "[they will be joining] together with colleagues" sub sentence. He is basical
95.
by DalasNoin 3y ago
I think the most aggressive part of this is the "together with colleagues" sub sentence. He is basically openly poaching the employees of a company that he supposedly has a very close cooperation with. This situation seems especia
96.
by DalasNoin 3y ago
I just realized this is from 2019 and confirmed this literally still works on PayPal. SMH
97.
by DalasNoin 3y ago
I tried to use it, but unfortunately it often did not add the little annotations for the different options to the screen and it got stuck in a loop. This bot works by adding a two letter combination to each clickable option, but sometimes t
98.
by DalasNoin 3y ago
HellaSwag and MMLU both improve slightly by about 1% but unsure if that is indicative of anything. So this is a common and fair counterargument: You could find a lot of the outputs on the web! Well, it certainly can't come up with real
99.
by DalasNoin 3y ago
I am the author of this paper. There was a post about a related lesswrong post before on HN https://news.ycombinator.com/item?id=37871203
100.
LoRA Fine-Tuning Efficiently Undoes Safety Training in Llama 2-Chat 70B(arxiv.org)
3 points by DalasNoin 3y ago|3 comments
101.
by DalasNoin 3y ago
My parents' cat has been castrated, what about your pets?
102.
by DalasNoin 3y ago
For amusement purposes, they should have never changed the old ms paint from windows XP and just added advanced AI to it. Same with clippy
103.
by DalasNoin 3y ago
Thanks for this, this narrative is being repeated a lot and should be challenged more: "The WAR on drugs failed so therefore we must totally 180 on everything." Obviously being very hard on drug dealers didn't eliminate drug
104.
by DalasNoin 3y ago
People are downvoting you. But 'pandering' to authority is a valid point if the other site does an ad hominen (These people are not technical, don't REalleY understand what they are talking about)
105.
by DalasNoin 3y ago
When using system prompts, you still need to finetune the language model to follow the content of the system prompt. I think that you can't quite replace constitutional AI with a constitution-systemprompt even if you do this. In practi
106.
by DalasNoin 4y ago
While I don't like tether and am aware of some of its shady practices, people who bet against it have lost a lot of Bayes points at this moment in time.
107.
by DalasNoin 4y ago
Have you tried just asking gpt4 to write a script that then implements the SAAS as a script that then calls gpt4? You should be careful about analogies, intelligence is special.
108.
by DalasNoin 4y ago
How about fentanyl and how about other theoretical substances with higher addictiveness and mortality rates?
109.
by DalasNoin 4y ago
Even if alignment works and AI is "safe", what does that actually mean? Frank Herbert in Dune wrote about AI: 1. "Once men turned their thinking over to machines in the hope that this would set them free. But that only permi
110.
by DalasNoin 4y ago
Clearly language models can already fool people into thinking they are human, we might be getting quite close to the adversarial turing test already. In the end, a good initial prompt might be the solution to this, something like "pret
111.
by DalasNoin 4y ago
Are you aware of some of the recent progress? Did you have a look at the Gato model and Flamingo by DeepMind, or at the chat logs of models like chinchilla and lambda? Or Alphacode? This is all from this year. I think your point is that all
112.
by DalasNoin 4y ago
You seemingly are portraying people who worry about long term risks of ai as members of a religious cult. But you also acknowledge that AI could end humanity? The question of why AI would want to kill us has been addressed by other people b
113.
by DalasNoin 4y ago
I think you are conflating different concepts. It is clearly imaginable that a very intelligent agent could end humanity if its objective would require so. How exactly is "perpetual motion machines can't exist" related to thi
114.
by DalasNoin 4y ago
Additionally to what others have mentioned, there is also the effect of the gravitational pull of ice shields. https://www.antarcticglaciers.org/question/antarctic-ice-she... There was a discussion about this effect on
115.
by DalasNoin 4y ago
I believe they only apply it during training.
116.
by DalasNoin 4y ago
On the other hand the openai tokenizer gives me a different tokenization ap - opl - oe [0]. If you capitalize A the result is A - pop - loe. The dalle 2 paper only specifies that it uses a BPE encoding, I would assume they used the same one
117.
by DalasNoin 5y ago
Scott Aaronson is featured quite frequently on Hackernews. I feel like many will trust him even if he speaks in an emotionally charged way. But that being said, the article contains one concrete example in the foreword by Scott and another
118.
by DalasNoin 5y ago
I assume the end goal of this is rendering photo-realistic movies of humans without filming actors. Neural voice generation is also quite good at this point. Quite interesting where this could lead to.
119.
by DalasNoin 5y ago
I think Jared Diamond told a number of convincing stories in this book to convince people of geographical determinism. However, I don't think that any of it amounts to a real proof. Essentially, he picks out more or less "successf
120.
by DalasNoin 5y ago
Is there evidence that monastic orders selected kids for intelligence? This effect should have then stopped with the protestant reformation in parts of europe. Maybe it is part of the reason why protestant heritage countries are more succes

More ›