Skip to main content

A Tree of Thoughts approach to LLMs suggests we have only scratched the surface of their power.

 



In a new paper by Long from Theta Labs 'Large Language Model Guided Tree-ofThought', Long hypothesises that there are two main contributing factors which limits the problem solving ability of LLMs:

Lack of long-term planning: LLMs are trained on massive datasets of text and code, but this data is typically organized in a linear fashion. This means that LLMs are not well-equipped to handle tasks that require long-term planning and strategic thinking.

Inability to explore multiple solutions: LLMs typically generate solutions to problems by following a single path. This means that they are not able to explore multiple possible solutions and choose the best one.

We believe that these two factors can be addressed by training LLMs on data that is organized in a more hierarchical fashion. This would allow LLMs to learn how to plan for the future and explore multiple possible solutions to problems.

So What is a Tree of Thoughts?

The Tree of Thoughts framework is a way of organising and representing knowledge that can be used to solve problems. It is based on the idea that all knowledge can be represented as a tree, with the root node representing the problem to be solved, and the leaves representing the possible solutions. The branches of the tree represent the different steps that need to be taken to solve the problem.

This framework can be used to translate classical insights about problem-solving into actionable methods for contemporary LMs. Classical problem-solving methods typically involve breaking down the problem into smaller subproblems, solving the subproblems, and then combining the solutions to the subproblems to solve the original problem. The Tree of Thoughts framework provides a way to do this in a systematic and organized way.

At the same time, LMs address a weakness of these classical methods, providing a way to solve complex problems that are not easily formalised, such as creative writing. Classical problem-solving methods are typically based on logic and reasoning, and they can be difficult to apply to problems that are not easily represented in this way. LMs, on the other hand, can be used to generate creative solutions to problems by exploring different possibilities and combinations of ideas.

Here is an example of how the Tree of Thoughts framework can be used to solve a problem. Let's say you want to write a story about a character who is lost in a forest. You could start by creating a tree with the root node "Lost in Forest." The first branch could be "Character is lost." The second branch could be "Character is scared." The third branch could be "Character is trying to find their way out." The leaves of the tree could be different possible solutions to the problem, such as "Character finds a path out of the forest," "Character meets a friendly animal who helps them find their way out," or "Character gets rescued by a search party."

Once you have created the tree, you can start to explore different possibilities and combinations of ideas. For example, you could start by generating a list of all the possible ways that the character could get lost in the forest. Then, you could start to brainstorm different ways that the character could find their way out of the forest. Once you have a few different ideas, you can start to flesh them out and write a story.


The paper from Long compliments another on ToT I've recently come across, 'Tree of Thoughts: Deliberate Problem Solving with Large Language Models', from: Yao, Yu, Zhao, Shafran, Griffiths, Cao and Narasimhan. 

What these two papers found, was somewhat remarkable: a 10x efficiency of answers for certain tasks. 

The deliberate Solving paper contains a warning at the end though, prior to the conclusion:

'ToT is a framework that empowers LMs to more autonomously and intelligently make decisions and solve problems. While current tasks are limited to reasoning and search problems, future applications involving interaction with external environments or humans could bring potential danger, e.g. facilitating harmful uses of LMs. On the other hand, ToT also improves the interpretability of model decisions and the opportunity for human alignment, as the resulting representations are readable, high-level language reasoning instead of implicit, low-level token values.'

It would seem, as tools like AutoGPT have already indicated, we are still only at an early stage of maximising the potential of the current set of LLMs that exist. 

Labels:

Comments

Post a Comment

Popular posts from this blog

The AI Dilemma and "Gollem-Class" AIs

From the Center for Humane Technology Tristan Harris and Aza Raskin discuss how existing A.I. capabilities already pose catastrophic risks to a functional society, how A.I. companies are caught in a race to deploy as quickly as possible without adequate safety measures, and what it would mean to upgrade our institutions to a post-A.I. world. This presentation is from a private gathering in San Francisco on March 9th with leading technologists and decision-makers with the ability to influence the future of large-language model A.I.s. This presentation was given before the launch of GPT-4. One of the more astute critics of the tech industry, Tristan Harris, who has recently given stark evidence to Congress. It is worth watching both of these videos, as the Congress address gives a context of PR industry and it's regular abuses. "If we understand the mechanisms and motives of the group mind, it is now possible to control and regiment the masses according to our will without their
Post a Comment
Read more

Beware the Orca, the challenge to ChatGPT and Palm2 is here

  So Google's 'we have no moat' paper was correct. If you train an LLM wisely then it's cost effective and cheap to produce a small LLM that is able to compete or even beat established, costly LLMs, as Microsoft has just found. It's another excellent video from AI Explained, who goes through some of the training procedures, which I won't get into here. Orca, is a model that learns from large foundation models (LFMs) like GPT-4 and ChatGPT by imitating their reasoning process. Orca uses rich signals such as explanations and complex instructions to improve its performance on various tasks. Orca outperforms other instruction-tuned models and achieves similar results to ChatGPT on zero-shot reasoning benchmarks and professional and academic exams. The paper suggests that learning from explanations is a promising way to enhance model skills. Smaller models are often overestimated in their abilities compared to LFMs, and need more rigorous evaluation methods. Explana
Post a Comment
Read more

What is happening inside of the black box?

  Neel Nanda is involved in Mechanistic Interpretability research at DeepMind, formerly of AnthropicAI, what's fascinating about the research conducted by Nanda is he gets to peer into the Black Box to figure out how different types of AI models work. Anyone concerned with AI should understand how important this is. In this video Nanda discusses some of his findings, including 'induction heads', which turn out to have some vital properties.  Induction heads are a type of attention head that allows a language model to learn long-range dependencies in text. They do this by using a simple algorithm to complete token sequences like [A][B] ... [A] -> [B]. For example, if a model is given the sequence "The cat sat on the mat," it can use induction heads to predict that the word "mat" will be followed by the word "the". Induction heads were first discovered in 2022 by a team of researchers at OpenAI. They found that induction heads were present in
Post a Comment
Read more