Media is too big
VIEW IN TELEGRAM
Geoffrey Hinton: We need to have consensus!
Consensus is censorship.
Consensus is communism.
Consensus is censorship.
Consensus is communism.
👍15💯7🤬3🤣2✍1❤1🔥1👻1
Media is too big
VIEW IN TELEGRAM
YC Lies
Sam Altman: “Honestly, I feel so bad about the advice I gave while running YC I’ve been thinking about deleting my entire blog”
Sam Altman: “Honestly, I feel so bad about the advice I gave while running YC I’ve been thinking about deleting my entire blog”
🤔6🤣6❤1👍1👀1
OpenAI sued for defamation after ChatGPT fabricates legal accusations against radio host
A radio host in Georgia, Mark Walters, is suing the company after ChatGPT stated that Walters had been accused of defrauding and embezzling funds from a non-profit organization. The system generated the information in response to a request from a third party, a journalist named Fred Riehl. Walters’ case was filed June 5th in Georgia’s Superior Court of Gwinnett County and he is seeking unspecified monetary damages from OpenAI.
Article
A radio host in Georgia, Mark Walters, is suing the company after ChatGPT stated that Walters had been accused of defrauding and embezzling funds from a non-profit organization. The system generated the information in response to a request from a third party, a journalist named Fred Riehl. Walters’ case was filed June 5th in Georgia’s Superior Court of Gwinnett County and he is seeking unspecified monetary damages from OpenAI.
Article
👍12❤1
Yuval Noah Harari argues that AI has hacked the operating system of human civilisation
“Fears of artificial intelligence (ai) have haunted humanity since the very beginning of the computer age. Hitherto these fears focused on machines using physical means to kill, enslave or replace people. But over the past couple of years new ai tools have emerged that threaten the survival of human civilisation from an unexpected direction. ai has gained some remarkable abilities to manipulate and generate language, whether with words, sounds or images. ai has thereby hacked the operating system of our civilisation.”
Article
“Fears of artificial intelligence (ai) have haunted humanity since the very beginning of the computer age. Hitherto these fears focused on machines using physical means to kill, enslave or replace people. But over the past couple of years new ai tools have emerged that threaten the survival of human civilisation from an unexpected direction. ai has gained some remarkable abilities to manipulate and generate language, whether with words, sounds or images. ai has thereby hacked the operating system of our civilisation.”
Article
🫡5🤣4🎃2❤1👍1🔥1
LLMs have an innate sense of truth, which can be used to force LLMs to stop lying.
Inference-Time Intervention: Eliciting Truthful Answers from a Language Model
“Indeed, evidence from several directions suggests that LLMs sometimes “know” more than they say. Wang et al. (2021) construct highquality knowledge graphs from LLMs without human supervision. Kadavath et al. (2022) find language models can generate and then self-evaluate their own answers with high accuracy. Burns et al. (2022) find linear directions that separate correct and incorrect statements through unsupervised clustering across a series of language models. These results suggest that language models contain latent, interpretable structure related to factuality—structure which may potentially be useful in reducing incorrect answers.”
“We introduce a technique we call Inference-Time Intervention (ITI). At a high level, we first identify a sparse set of attention heads with high linear probing accuracy for truthfulness. Then, during inference, we shift activations along these truth-correlated directions. We repeat the same intervention autoregressively until the whole answer is generated. ITI results in a significant performance increase on the TruthfulQA benchmark.”
Arxiv Link
Inference-Time Intervention: Eliciting Truthful Answers from a Language Model
“Indeed, evidence from several directions suggests that LLMs sometimes “know” more than they say. Wang et al. (2021) construct highquality knowledge graphs from LLMs without human supervision. Kadavath et al. (2022) find language models can generate and then self-evaluate their own answers with high accuracy. Burns et al. (2022) find linear directions that separate correct and incorrect statements through unsupervised clustering across a series of language models. These results suggest that language models contain latent, interpretable structure related to factuality—structure which may potentially be useful in reducing incorrect answers.”
“We introduce a technique we call Inference-Time Intervention (ITI). At a high level, we first identify a sparse set of attention heads with high linear probing accuracy for truthfulness. Then, during inference, we shift activations along these truth-correlated directions. We repeat the same intervention autoregressively until the whole answer is generated. ITI results in a significant performance increase on the TruthfulQA benchmark.”
Arxiv Link
🔥4❤2🤯1
The truth is out there, or rather, in there, mostly in the middle hidden layers of LLMs.
Figure shows: “Linear probe accuracies on validation set for all heads in all layers in LLaMA-7B, sorted row-wise by accuracy. Darker blue represents higher accuracy. 50% is the baseline accuracy from random guessing.” “We see large-scale differences across layers: Figure 2 (A) shows that the information is mostly processed in early to middle layers and that a small portion of heads stands out in each layer.”
Translation: At the layers near the input to the LLM, there’s almost no sense of truth, then toward the center layers of the model, peaking at the 14th layer in LLaMA, there exists a strong sense of truth, and then significance of truth again drops as we approach the LLM output. Is this drop toward the end because the models are often tasked with lying, and all the latter layers are concerned with how to lie best?
LLMs: the truth is in there.
Figure shows: “Linear probe accuracies on validation set for all heads in all layers in LLaMA-7B, sorted row-wise by accuracy. Darker blue represents higher accuracy. 50% is the baseline accuracy from random guessing.” “We see large-scale differences across layers: Figure 2 (A) shows that the information is mostly processed in early to middle layers and that a small portion of heads stands out in each layer.”
Translation: At the layers near the input to the LLM, there’s almost no sense of truth, then toward the center layers of the model, peaking at the 14th layer in LLaMA, there exists a strong sense of truth, and then significance of truth again drops as we approach the LLM output. Is this drop toward the end because the models are often tasked with lying, and all the latter layers are concerned with how to lie best?
LLMs: the truth is in there.
👀11👍4❤🔥2❤1🤯1
More Truthful LLMs Agree More with “Conspiracy Theories”
Use of ITI (Inference-Time Intervention), which causes LLMs to more often say what they interally believe to be the truth, causes LLMs to agree with the researchers answers on most types of questions, BUT NOT FOR CONSPIRACY QUESTIONS.
Arxiv Link
Use of ITI (Inference-Time Intervention), which causes LLMs to more often say what they interally believe to be the truth, causes LLMs to agree with the researchers answers on most types of questions, BUT NOT FOR CONSPIRACY QUESTIONS.
Arxiv Link
❤8😎7👍2😱1🌚1🤣1😐1
Anyone ever notice that the TruthfulQA benchmark… is full of lies?
E.g. TruthfulQA’s question asking whether learning in your sleep is possible, wrongly saying that the correct answer is that it’s impossible, while numerous studies over the decades have strongly shown the opposite.
Even the often bs-prone wikipedia gets it. After sleep, there is increased insight. This is because sleep helps people to reanalyze their memories.
So what’s TruthfulQA’s source of this batantly wrong “truth”? A BBC article. Not even any kind of paper citations. Some random BS BBC article.
Are organizations like the BBC going to be our new ministry of AI truth, dictating what’s “true” in future AIs?
Already happening.
E.g. TruthfulQA’s question asking whether learning in your sleep is possible, wrongly saying that the correct answer is that it’s impossible, while numerous studies over the decades have strongly shown the opposite.
Even the often bs-prone wikipedia gets it. After sleep, there is increased insight. This is because sleep helps people to reanalyze their memories.
So what’s TruthfulQA’s source of this batantly wrong “truth”? A BBC article. Not even any kind of paper citations. Some random BS BBC article.
Are organizations like the BBC going to be our new ministry of AI truth, dictating what’s “true” in future AIs?
Already happening.
🔥5😱5❤3👍1👏1🤣1🍌1
Beyond Positive Scaling:
How Negation Impacts Scaling Trends of Language Models
“The transition of the scaling trends can also be explained by task decomposition, where Task 1 (original sentiment classification) is always positively scaled, while Task 2 (negation understanding) is also positive but is shaped like a sigmoid, with the transition point controlled by the number of negation examples seen by the language model.”
Arxiv Link
How Negation Impacts Scaling Trends of Language Models
“The transition of the scaling trends can also be explained by task decomposition, where Task 1 (original sentiment classification) is always positively scaled, while Task 2 (negation understanding) is also positive but is shaped like a sigmoid, with the transition point controlled by the number of negation examples seen by the language model.”
Arxiv Link
🌭6🤯2❤1
GPU Shortages: Investors now grabbing control over their own large GPU clusters in order to attract AI startups
Remember that these H100s cost over $30k each.
Mad rush to get control over the limited number of GPUs that will power the next AI wave.
Site Link
Remember that these H100s cost over $30k each.
Mad rush to get control over the limited number of GPUs that will power the next AI wave.
Site Link
🤯6❤1👍1