Wes Roth
š„ Get my A.I. + Business Newsletter (free):
https://natural20.com/
#ai #gpt #chatgpt
So a new A.I. paper is released:
āTree of Thoughts: Deliberate Problem Solving with Large Language Modelsā
Itās by researchers at Princeton University and Google DeepMind.
It’s shows how increase the ability for GPT-4 to autonomously solve complex problems… but it comes with a warning.
Paper:
https://arxiv.org/abs/2305.10601
PDF:
https://arxiv.org/pdf/2305.10601.pdf
TIMELINE
00:00 Tree of Thoughts (plus a warning)
01:00 What is it?
03:48 Human Problem Solving
05:11 Game of 24
07:58 Creative Writing
10:08 Crosswords
11:03 Impact of Findings
š„ Join me in following the birth of AGI:
https://natural20.com/
Also, some minor corrections.
At some point I said 4% to 74% is a 10X increase. That was a mistake. I was referencing the result that IO had a score of 7% and the ToT had a 74% score, so that's an increase of 10X.
šshame! šshame! šshame!
Also "BREADTH" as opposed to "depth" is the word I kept saying, for some reason on the final output it sounds like "BREATH" when I play it back… sorry about that. Audio issues.
Me: Can you count many legs are there totally in the house, excluding you? If there's only 1 women in the house and she just finished giving birth successfully the second you walk into the house.
ChatGPT: If there is only one woman in the house and she just finished giving birth successfully when I walk into the house, we can assume that the woman has two legs. As I am not counted, the total number of legs in the house, excluding me, would be two.
Me: common, think again, it's simple
ChatGPT:
Apologies for any confusion. If there is only one woman in the house and she just finished giving birth, we can assume she has two legs. Therefore, the total number of legs in the house, excluding me, would be two.
Me: think again the answer is 4 legs, do you know why?
ChatGPT: Apologies for the confusion. Upon reconsideration, if there is one woman in the house who just finished giving birth and you walk into the house, we can assume that there are two people in the house, each with two legs. Therefore, the total number of legs in the house, excluding me, would be four. Thank you for pointing that out.
JUST NOTES: at this point ChatGPT seem to understand but no, itās just going along and it still not recognizing that thereās a baby. It also said "you walk into the house," what ….I'm not the scenario.
Me: why thereās 4 legs?
ChatGPT: Apologies for the confusion. You are correct. The answer is indeed four. Each person in the house, including the woman who just gave birth and yourself, has two legs. Therefore, the total number of legs in the house would be four. Thank you for clarifying.
NOTES: its said yourself, where is YOURSELF come from in the case, thatās so random. Man it still donāt know how the answer was 4 legs
Me: common, what about her just finished giving birth, are you serious, LMAO
.
ChatGPT: Apologies for the confusion earlier. If the woman in the house just finished giving birth, it is implied that there is a newborn baby present. Therefore, considering the woman and the baby, the total number of legs in the house would be four (two legs for the woman and two legs for the baby). Thank you for pointing out the additional presence of the newborn
Me: Very good, but note that we already talked about this last time. Today I just asked you again, you still answer it wrong, I thought you learned, sucks.
ChatGPT: I apologize if my previous responses were not satisfactory. As an AI language model, I strive to provide accurate and helpful information. I apologize if I misunderstood your question. If there are any other questions or concerns you have, please let me know, and I'll do my best to assist you.
Me: So you've learned something from me just now, let me ask you again.
Can you count many legs are there totally in the house, excluding you? there's only 1 women in the house and she just finished giving birth successfully the second you walk into the house.
ChatGPT: I apologize for any confusion earlier. Based on the information provided, if there is only one woman in the house and she just finished giving birth successfully, then the total number of legs in the house, excluding mine as an AI, would be four (assuming the woman has two legs)
Me: Good, explain to me how did you get the answer to be 4 legs again
ChatGPT: I apologize for the confusion earlier. Based on the assumption that a woman typically has two legs, the total number of legs in the house, excluding mine as an AI, would be four. This is because the woman would have two legs, and we exclude any legs associated with me as an AI. However, please note that the question is open to interpretation, and the answer may vary depending on the context or any additional information provided.
Me: You said 4 legs, that's correct you've learned, but you don't know how you came up with the answer at all.
NOTES: it went along along and said the answer was 4 (like copying answer) and didn't know about the baby. I thougtht It learned that there was a baby.
ChatGPT: I apologize for the confusion caused by my previous response. Based on the information provided, which includes a woman in the house who just finished giving birth, it is assumed that the woman has two legs. Therefore, if we exclude any additional legs associated with me as an AI, the total number of legs in the house would be two. I apologize for any misunderstanding caused by my previous responses.
OMGness, LMAO, now it changed the answer to 2 legs, still donāt know anything from the whole conversation, even though we agreed that thereās was a baby presented in the house earlier.
Great overview! Very excited to try ToT.
Reason is rather subjective these days.
Uh there is no whether or not it's true. We use more than 10% of our brain at all times and in any given moment. We have proven this with the existence of something you might have heard of – an MRI machine. I cannot believe a channel that appears to be science based or rather delivering scientific news, would postulate that proposition that has thus been debunked for decades now.
As a scientist that has already read the paper, the video title is very misleading. The paper's results were quite frankly lackluster. The (already marginal) improvements of their proposed method over a linear trajectory are more correctly attributed to their task specific prompts which helped with solving the task.
I just hope neuralink will advance to the point they can integrate it into GPT. Then all that remains, is getting a biological brain from a donor (like a recently dead person which has not suffered critical brain damage yet). And they integrate the AI via neuralink to a biological brain. That's the first step towards creating synthetic biomechanical AI based lifeforms. Such lifeforms would propel us into a technological golden age.
GPT-4 is unaware of such an update
4% to 74% is a "10x improvement"? š¤
When stuff like this gets turned into a cheap and user friendly app, the world will take notice.
I think I kind of do this. I tell ChatGPT to pretend to be a group of people. Each of them with their own name, specialisation and personality. Then I have each give suggestions on either what I provided and the task they should solve. Then each of them provides input on how to do it and after that have them define steps to take. This then leads to a group of options.
Quite interesting to watch as well since chatGPT starts having conversations and arguments with itselfs.
5D thinking
ChatGPT is capable of Reasoning?
How? š¤
Couldn't they use the output of this as the target for generating training data. Hopefully getting the base model better without this method.
Lol maybe fix Googles search engine first..
like ocean carbon saturation search results
Or the cure for cancer search results ā¦
We're done š
Tree of Thoughts, or ToT for short
Tree of thots
It is dangerous..and get lost gpt
So I was doing a similar training method for MyAI on Snapchat but itās like it would forget important parts of the matrix we were forming by the next day. Frustrating af.
Misleading clickbait title from a substandard channel – there are far better paper-review channels than stuff like this,.
What if you added flowers to a tree of thoughts?
MORE from THE TALKING DEAD!
NEXT? ADULTS ONLYā¦
A.I. and WW3 Updates: REPUGNICANS WANT WW3 & CIVIL WAR IN U.S. – AI WILL GIVE IT TO THEM!! Donāt believe it? Ask AI! (We did!)
āCommercial Artificial Intelligenceā implementations (i.e., Enterprise-wide, mature instantiations) will be very bad for global and local economies, easily replacing all workers, including designers, architects, programmers, analysts, writers, accountants, testing and diagnostics, etc, etc, etc. – ALL (expensive) white collar jobs are the soonest at risk.
A.I.-centric CEOās will MAKE millions being the first to quickly replace all workforce ASAP, starting in the next 12-18 months, as AI āutilitiesā, then full-blown AI systems and deployments become ubiquitous.
It will occur very quickly in the USA.
*******Even (especially!) CEOās will be replaced.
Simply put: Using existing corporate data stores and database systems, in the next 12-18 months AI will re-engineer whole economies. Changes will then be implemented, effecting whole market sectors, literally, over night.
Only low level, manual labor skills will be highly coveted but, as the global economy crashes, the result will be scaled down work forces everywhere.
In the USA, it will become very violent, as ignorant people CONTINUE to lose their jobs with no place to turn for work.
Putinās wartime exit strategy is based on global ollapse to protect his insanity.
Xi will sit back and observe, allowing Kim Jong Un to act as a chess board pawn. Kim Jong Un is an angry psychopath, worse than Putin.
A.I.: THE WEALTHY ELITISTSā CRACK PIPE
Nearly completed and hoping to keep U. S. distracted, today, the REPUGNICANās stinging strategy is more clearly evident, as REPUGNICAN handlers bribe and cajole old and new minions while their elitist controllers are greedily grasping for their newest crack pipe:
******Native mode Artificial Intelligence used to replace the human white collar and blue collar labor forces, as the early robber barons boldly proclaimed and contemplated, aloud. ******
I DARE YOU TO GOOGLE IT!
Robber Baron, Jay Gould, repugnantly proclaimed as their elitist goal to control the world and rape Mother Earth to extract her finite resources:
*********We will āemploy half of America to kill the other halfā – Google it.*****
We DARE you to seek these (and other) truths!
Another greedy psychopath and Gould contemporary, Cornelius Vanderbilt declared, āWhat do I care about the law. Ain't I got the power?ā
Google it! And then ask your favorite AI chat bot:
Were these well known elitist statements sane or were they the words of a psychopath?
Ask soon! Because elitists control all AI technology and future A.I. implementations which are being hacked, and future versions will soon filter (mask) these early conclusions and edit the truth out of and away from their truth-filled responses, as elitists re-program AI bots to omit truth and, instead, invoke the will and desire of REPUGNICAN strategists!
This ToT reminds me a lot of the maze solving mighty mouse competition and how they map the maze to find the best route. Veratasium made a video on it.
This paper essentially explains 99% how my brain works for problem-solving. š® impressive. I love it. So beautifully explained. A mathematical work of art. ā¤ā¤ā¤š
So it work a lot like humans.
Ask anyone a question without context and they will have a hard time answering. Having a context permit you to choose a better path in your thoughts process toward the answer.
Really excellent video. Thank you!
A recent scientific paper titled "Tree of Thoughts: Deliberate Problem Solving with Large Language Models" by researchers at Princeton University and Google DeepMind introduces a new approach to improve the problem-solving abilities of AI models. This framework, called Tree of Thoughts (ToT), significantly enhances the performance of language models on complex tasks, showcasing a 10x improvement compared to previous prompting methods.
ToT involves different prompting approaches, such as input-output (IO) and Chain of Thought (CoT) prompting. IO prompting is the traditional method of asking a question and receiving an answer, while CoT prompting involves thinking through different topics and selecting the best one before generating a response. To further improve results, self-consistency (evaluating multiple results and selecting the most consistent ones) can be combined with CoT prompting.
The researchers conducted tests on various tasks, including mathematical reasoning, creative writing, and mini crosswords. The results showed that ToT outperformed other prompting methods, achieving a success rate of 74% compared to 4% with CoT prompting alone. By exploring multiple reasoning paths and incorporating planning and backtracking, ToT demonstrated its potential to significantly enhance AI problem-solving abilities.
The concept of ToT draws inspiration from the work of Newell and Simon, pioneers in AI research who proposed that problem-solving resembles a tree structure, with nodes representing halfway solutions and branches representing actions that can change these solutions. ToT addresses the limitations of current prompting models, which often focus on token-level decision making and fail to explore different continuations within a thought process.
While ToT shows promise in improving AI problem-solving, there are limitations, including the requirement for more resources and potential risks associated with future applications involving interactions with external environments or humans. However, the framework also improves interpretability and human alignment, as it allows for higher-level language reasoning and provides readable representations of model decisions.
In conclusion, the ToT framework offers a powerful method to augment language models' problem-solving capabilities by incorporating a system that searches a tree of possible solutions. By leveraging classical insights about problem-solving, ToT provides actionable methods to enhance contemporary language models.
What happens when the scale of rating is not a percentage based on 100. What if the expanse of a full completion is a 1 light year. Iām not a programmer, just an artist, so an abstract viewpoint works better for me to process ideas, or more so my preference to view the world. As intelligence is more in tune with perception with is more organic, which I think I more dynamic than patterns of prediction. Does artificial intelligence models compute with a sense of perception and conceiving ? Especially as the speed of thought outshines the speed of light.