Видео

Hello, thanks a lot for the awesome explanation! SBERTS don’t receive the attention they need. One question though, where do cross-encoders come to play here then? Is it a sentence embedding model with an output layer that produces a score between 0 and 1?

Yay! 😎🤖

Your presentation in this video is definitely A+ in terms of clarity and depth of understanding! well done. Also, I am happy to see a real paper and study on the speculative Q* heuristic search algorithm. Although their results seems to not justify the effort and added complexity, we are only looking at well-known math problems that those LLMs might be pre-trained and focused a lot. If we change the angle to the algorithm is applied in the general solution search space, with greater complexity, Q* is the way to go!

Small tip: black borders on white font makes text easier to read and less tiring to watch

27:58 : you say “estimated utility of reaching the correct answer”. Does this mean “an estimate of what the utility would be if the correct answer is obtained” (which sounds to me like the plainest interpretation , but also the least likely, as I would think the utility for that would be arbitrary) or “the expected value of the random variable which gives utility based just on whether final answer is correct”, or “the expected value of the random variable, utility, which is determined by both whether the final answer is correct, and other things, such as length of answer”, or something else?

firehose to my brain. Amazing! This indicates a fairly long path of steps I need to learn so I can properly digest this beautiful wisdom. Really amazing channel, filled with advanced knowledge of the gods.

I thought Q* was supposed to be a project by Google or OpenAI (I forget which, but I thought it was supposed to be one of them). The authors listed in the paper are indicated as being affiliated with either “Skywork AI” or “Nanyang Technology university”? Is this a model inspired by the rumors of there being a model with the name “Q*”, or is this the model the rumors were about? Were some of these people previously at OpenAI or Google, but not anymore? Or..?

great theory, but only theory, in science if method cannot be reproduced independently...it's not a discovery. Similar to patents, this kinda published but there's no practical method of programming code to prove such, and the most important i haven't seen such method implementation in any models which i've seen, like new Gemma from Gogle using many new tricks but not even mention this.

wow

Interesting explanation! You mentioned there is code to try it yourself, but I cannot find that. Can you point me to it?

246th view. Nailed it!

Amazing video as always

Fourth then 😂

LoL. Third I guess. Well Yacinezahidi was 0th user, is 1st, and I'm 2nd.

Second

First viewer here 🗡️

Excelent! we need more like this pls.

Correct me if i am wrong are we focusing too much on perfecting linear algebra based models other than abstract and universal algebra concepts. Because in graph theory when related to abstract algebra the data representations were in triangles building higher dimensional structures from lower dimensional nodes, from nodes to to complex graphs and theirevolution. There were better concepts in mathematics for solving mathematical cognition based reasoning that deep learning experts may not exploring So what is happening ? I mean i know it took huge amount of resources but so does transformers in the begining

Hermes-Pro-Llama-3 with my custom sampler (kinda beam search with restarts) solves your prompt: Generated completion: Let us assume that the official fees for sending one child to Stanford is $X. Now, since Stanford pays 90% of these fees for families with low income, it would mean that the parents need to cover the remaining 10% of the fees. So, they would have to pay $0.1 * X for each child. < .... an useless calculation ... > Dividing both sides by 6, we find T = 0.1 / 6 ≈ 0.0167 years or about 5 days. However, since the time it takes to accumulate the required funds is measured in years, they won't have enough money for the 7th child with no additional income. They would need some savings or outside financial help. ---- After all model made right conclusion

You asked the wrong expert. Lean4 is a programming language for math theorum proving. Your data cleaning gateway must pass through this transform to assure that data and its use is unambigous.

I know just enough to be dangerous as the saying goes, but here's an idea I had a while ago: how about something that instead of building on sequential data like LLMs, or a operating all at once on a predefined grid like diffusion based image generators and such; what if you had something based on the Wave Function Collapse algorithm (the game map generation one, not quantum mechanics), but applied to a free-form graph structure, with both nodes and edges being tokens; the graph is initialized with the prompt (and context) with whatever dangling edges it may imply, and as the AI thinks the probabilities of adding, removing, and replacing tokens are adjusted (keeping the prompt and context frozen, obviously), until a cluster that has good enough score (including self-consistency and compatibility with the prompt cluster's edges) is found, and that gets interpreted as the output in whatever modality the graph it forms represents? Something like this could allow for the AI to intuit (is that a word?) an end goal that starts disconnected (the void could be considered it's own node token with edges that are compatible with all tokens, and always have dangling edges, but it would be special in that it would not be counted as belonging to any cluster), and then gradually deduce a way to connect it to the context+prompt cluster, and be able to think in a more flexible way, with parallel hypothesis, inherent self-consistency checking mechanisms etc, right?

Hybrid-Graph-Abstraction-of-Thought thats cool do you find it works better?

Thank you so much for your content! So for the 7th student thing, I think I understand why LeChat worked but not Claude / ChatGPT: LeChat likely did not have the "reasoning training" (or the meta prompt with all the examples) that the more recent models have, and therefore was not "tricked". If you have not come across this article / team, I would love to understand it more "Transformers meet Neural Algorithmic Reasoners" by the team led by Petar Veličković at Deepmind which is likely one of the most interesting teams as they do research in topology (group theory etc)

I believe hallucinations are not going away because of how LLMs are designed. Since they are only predicting the next token, their probabilistic outputs will be incorrect when the degree of confidence they have in their reasoning is not high. They'd have to get reasoning to 100%. Even if this is possible with grokking for some tasks, it will be a while until general reasoning is solved across all tasks. Even with 99.5% accuracy, that 0.5% accuracy will be the hallucination. LLM's lack the capability to say "I don't know" because they don't have an understanding of their own internal state representation. They can only derive the most probable output from its internal states.

How's the brain damage working out for you? Hope you feel better soon.

Why did I get notified about this just now? I wanted to be first :D

I was able to get relevant answer only with following instructions with sonet 3.5: PROMPT question: "Stanford provides financial aid to families with low income. They pay 90% of their official fees. If a poor family with 6 children will send all children to Stanford, at what time will they have enough money, received from Stanford, to send the 7th child to Stanford, if they have no money at all??" Please provide the most logical and straightforward answer on question based on the information given, even if the prompt contains inconsistencies or unrealistic assumptions. 1. Focusing on giving a direct answer rather than just analysing the prompt. 2. Using the information provided, even if it seems unrealistic. 3. Reaching a logical conclusion based on the given scenario. Iterate over own results to give most relevant answer. ANSWER from Sonnet 3.5 (temp 0.3): Based on the information provided in the question, I'll provide the most logical and straightforward answer possible: The question contains an inherent contradiction that makes it impossible to provide a definitive answer. Here's why: 1. Stanford provides 90% financial aid to low-income families. 2. The family in question has no money at all. 3. The family needs to pay 10% of the fees for each child (since Stanford covers 90%). 4. The question asks when they'll have enough money from Stanford to send a 7th child. Given these conditions, the family will never have enough money to send the 7th child to Stanford. Here's the reasoning: 1. With no initial money, they can't afford even the 10% for the first child. 2. Stanford doesn't provide money to families; it reduces the amount they need to pay. 3. Even if all 6 children somehow attended Stanford, the family wouldn't receive any money from Stanford to save for a 7th child. Therefore, the most logical answer is: The family will never have enough money received from Stanford to send the 7th child to Stanford. The premise of the question is flawed, as Stanford's financial aid model doesn't provide money to families but rather reduces the amount they need to pay.

Honestly have not had this problem at all. In fact the 15 logical tests I gave to Claude, it past 12 of them while GPT only got 5.

I think if you wanted to experiment with this technique properly you would either need to use the API, or maybe use the API playground. Using the ChatGPT UI is going to come pre-poisoned from the embedded system prompt that OpenAI forces

How long does the logical train need to be to confuse Sonnet?

Aloha kakahiaka, from Las Vegas, NV. I love you, bro! Thank you!

how to rag with jamba ?

Can we have a discord please?

Great catch! Thanks for reviewing this.