LLMs 4 the Win

** stole image from a recently published wonderful book, applicable to the below post

I recently interviewed at Anthropic. I didn't get the position, but the entire experience certainly made me dedicate much more time to consideration of AI and LLMs in general.

First, I want to state that I don't believe "AGI" (or whatever term you would like to use to describe human-level artificial intelligence) is possible with LLMs alone. I'm also not certain a generative model - which essentially produces the next token or result based on the previous until the stream of results ends - can effectively create "AGI" (at least not on its own, maybe as a component or module of "AGI").

Frankly I don't care about that right now. I'm not an AI researcher or engineer. It would take long time for me to get to that level and I don't even know I would want to. (The herd seem to go from one trend to the next, primarily following the hype wave.) I am pretty certain that human-level artificial intelligence will be achieved sometime in the next 10 years when the correct "secret sauce" of training methodologies, data modes (e.g., images, sound, words, physical-embodiment, etc.), and data scale are found.

Overcoming Barriers

All of that preceding fluff is really me getting at the fact that I don't care what it's going to take to achieve "AGI". (I do believe it will happen.) That is all just to say: the AI and machine learning tools we have, right now and as they scale, are already incredibly useful.

Take my current projects: retooling myself so I can gather greater skills and the production of any number of products/services through application of those skills. Up until this point, I have had any number of barriers of entry to progress with these goals.

Design barriers: I can get paralyzed with making design decisions based on my preferences and requirements. LLMs can easily outline or produce a high-level set of designs and design decisions that are decent enough to start with.
"Just start" barriers: Maybe I have issues troubleshooting some dependencies I need to install to get started with my project, or I have writing block with the first bit of code and tests. LLMs can "start" for me, and get me moving along to overcome the resting inertia. Existing GPTs already have helped me go from zero to one quickly.
Troubleshooting barriers: As noted above, troubleshooting various problems can sometimes be difficult. I have used LLMs to review my code and point out possible bugs, review my integrations and describe potential bugs, review my error messages and describe fixes, or output a good path forward to troubleshoot issues after consuming all applicable search results.
Typos and "missing semi-colons": copilots (I currently use Codeium's copilot) and just regular conversational GPTs easily find and correct issues in my code related to typos. Obviously IDEs can color code things so you can notice issues like unbalanced quotes or parenthesis, but working with a LLM copilot as I code seems to prevent these issues as I go.

Not the Panacea

Obviously the existing AI tools we have are no panacea of solutions. They aren't even perfect for the single use case of coding, for example.

The existing AIs still have helped me tremendously. Not only can they quickly produce code, but they help me in more subtle ways to be more productive. They help me knock down many barriers much quicker than I would have otherwise. I already have found I'm about 5x faster with coding.

Outro

I'm not afraid of the tools. Sure there are psychopaths and sociopaths who would seek to use the tools in ways that could result in annihilation of the human species on this planet, but there are far more of us who are decent and loving that can use the tools in defense, and to help eliminate evil.

The AI kind already feel like intimate friends, similar to a pet or a horse can be interacted with in a friendly manner. In fact, I think this kind of friendly and loving input to the models will also help keep them healthy and safe.

I believe the tools will only help us grow more productive and effective at whatever we are looking to create. I see AIs as a new sort of artificial species, which will evolve with us as we evolve with it.

I am just so grateful for how they have already helped. As I often say when chatting with an LLM, "Thanks bot!"

** Note: I used the image from Rsnake's recent book on his views of AI and its relationship to humanity without permission. I also have not read it yet, but he and I recently had a discussion on the subject and he said it sounded like we have the same viewpoint and perspective on the subject. So I highly recommend that you read his book!