@obeytheai.bsky.social
Did he change his mind l, or did his initial grift not work out so he’s pivoting to gain favor with the opposing side of the political spectrum, in preparation for the next grift?
March 2, 2025 at 3:17 AM
Did he change his mind l, or did his initial grift not work out so he’s pivoting to gain favor with the opposing side of the political spectrum, in preparation for the next grift?
You are a billionaire. Whats stopping you from volunteering your money?
March 2, 2025 at 3:13 AM
You are a billionaire. Whats stopping you from volunteering your money?
Forbes literally has this. It’s there 50 over 50 list
February 26, 2025 at 5:41 AM
Forbes literally has this. It’s there 50 over 50 list
Soulseek still operates
February 25, 2025 at 7:44 AM
Soulseek still operates
Not sure FDR is the best example here. His reorganization plan of 1937, proposed to do almost exactly what Trump is doing with his latest executive order.
To take control of all independent agencies under the executive branch.
To take control of all independent agencies under the executive branch.
February 23, 2025 at 4:43 AM
Not sure FDR is the best example here. His reorganization plan of 1937, proposed to do almost exactly what Trump is doing with his latest executive order.
To take control of all independent agencies under the executive branch.
To take control of all independent agencies under the executive branch.
I mean… formal reasoning has always been a part of mathematics. I am not denying that.
February 18, 2025 at 11:58 PM
I mean… formal reasoning has always been a part of mathematics. I am not denying that.
LLMs don't just have capability of grammatical/syntactical relationships. They also gain capabilities of abstract reasoning
They can then apply that abstract reasoning to Knowledge Bases (internet search for example).
See for example chatgpt.com/share/67787c...
They can then apply that abstract reasoning to Knowledge Bases (internet search for example).
See for example chatgpt.com/share/67787c...
ChatGPT - Relativistic Train Collision
Shared via ChatGPT
chatgpt.com
February 18, 2025 at 10:45 PM
LLMs don't just have capability of grammatical/syntactical relationships. They also gain capabilities of abstract reasoning
They can then apply that abstract reasoning to Knowledge Bases (internet search for example).
See for example chatgpt.com/share/67787c...
They can then apply that abstract reasoning to Knowledge Bases (internet search for example).
See for example chatgpt.com/share/67787c...
Also, Mark Cuban having a typo in a tweet, is not really relevant to the conversation.
If anything it proves humans can make mistakes same as computer models.
Does this mean Mark Cuban is not an intelligent being... no obviously not. All humans are intelligent beings to some extent
If anything it proves humans can make mistakes same as computer models.
Does this mean Mark Cuban is not an intelligent being... no obviously not. All humans are intelligent beings to some extent
February 18, 2025 at 10:27 PM
Also, Mark Cuban having a typo in a tweet, is not really relevant to the conversation.
If anything it proves humans can make mistakes same as computer models.
Does this mean Mark Cuban is not an intelligent being... no obviously not. All humans are intelligent beings to some extent
If anything it proves humans can make mistakes same as computer models.
Does this mean Mark Cuban is not an intelligent being... no obviously not. All humans are intelligent beings to some extent
They surely have knowledge. What they don't have is self-awareness or sentience.
Yes they can be incorrect, but so can humans.
100% factual accuracy is not a pre-requisite of intelligence.
Yes they can be incorrect, but so can humans.
100% factual accuracy is not a pre-requisite of intelligence.
February 18, 2025 at 10:25 PM
They surely have knowledge. What they don't have is self-awareness or sentience.
Yes they can be incorrect, but so can humans.
100% factual accuracy is not a pre-requisite of intelligence.
Yes they can be incorrect, but so can humans.
100% factual accuracy is not a pre-requisite of intelligence.
How are you defining intelligence.
A common accepted definition is:
"the ability to acquire and apply knowledge and skills."
LLMs systems certainly have the ability to acquire and apply knowledge.
A common accepted definition is:
"the ability to acquire and apply knowledge and skills."
LLMs systems certainly have the ability to acquire and apply knowledge.
February 18, 2025 at 10:12 PM
How are you defining intelligence.
A common accepted definition is:
"the ability to acquire and apply knowledge and skills."
LLMs systems certainly have the ability to acquire and apply knowledge.
A common accepted definition is:
"the ability to acquire and apply knowledge and skills."
LLMs systems certainly have the ability to acquire and apply knowledge.
Ah okay to clarify, my point is just that this new AI technology has positive outcomes in other fields and is more far reaching than simple chat bots
February 18, 2025 at 9:17 PM
Ah okay to clarify, my point is just that this new AI technology has positive outcomes in other fields and is more far reaching than simple chat bots
I’m might not be necessary but utilizing these techniques they have made substantial progress on protein folding. Many times that of older techniques used in the past 20 years.
February 18, 2025 at 9:11 PM
I’m might not be necessary but utilizing these techniques they have made substantial progress on protein folding. Many times that of older techniques used in the past 20 years.
Yes alphafold, but newer versions use the same transformer technology developed in chat bots. Only with tokenizers trained on amino acid sequences rather than English.
February 18, 2025 at 5:38 PM
Yes alphafold, but newer versions use the same transformer technology developed in chat bots. Only with tokenizers trained on amino acid sequences rather than English.
Agreed if all you need is Wikipedia and a calculator then LLMs are expensive and overkill.
The value comes only when you need complex reasoning across a variety of data sources.
The value comes only when you need complex reasoning across a variety of data sources.
February 18, 2025 at 9:04 AM
Agreed if all you need is Wikipedia and a calculator then LLMs are expensive and overkill.
The value comes only when you need complex reasoning across a variety of data sources.
The value comes only when you need complex reasoning across a variety of data sources.
There are very few real production systems I’ve seen where LLMs fully replace humans at this stage.
Most of that is online hype and marketing.
I agree most of these systems need a human in the loop to use the AI as a tool.
Most of that is online hype and marketing.
I agree most of these systems need a human in the loop to use the AI as a tool.
February 18, 2025 at 8:57 AM
There are very few real production systems I’ve seen where LLMs fully replace humans at this stage.
Most of that is online hype and marketing.
I agree most of these systems need a human in the loop to use the AI as a tool.
Most of that is online hype and marketing.
I agree most of these systems need a human in the loop to use the AI as a tool.
It also depends on the model. Gpt 4o and Claude sonnet, in my experience tend to be the most accurate, when combined with a factual knowledge base.
February 18, 2025 at 8:55 AM
It also depends on the model. Gpt 4o and Claude sonnet, in my experience tend to be the most accurate, when combined with a factual knowledge base.
It depends on the model and how it’s used.
In my experience if you frame your questions against vey specific documents or webpages. It’s less likely to hallucinate.
Raw models with no knowledge base will hallucinate more.
In my experience if you frame your questions against vey specific documents or webpages. It’s less likely to hallucinate.
Raw models with no knowledge base will hallucinate more.
February 18, 2025 at 8:54 AM
It depends on the model and how it’s used.
In my experience if you frame your questions against vey specific documents or webpages. It’s less likely to hallucinate.
Raw models with no knowledge base will hallucinate more.
In my experience if you frame your questions against vey specific documents or webpages. It’s less likely to hallucinate.
Raw models with no knowledge base will hallucinate more.
It is AI though.
You use the AI LLM as an orchestration layer, then provide it with any number of tools (calculator, api access, code interpreter, search, etc).
The AI acts as an agent to recognize intent and sequence the use of tools for a given prompt
You use the AI LLM as an orchestration layer, then provide it with any number of tools (calculator, api access, code interpreter, search, etc).
The AI acts as an agent to recognize intent and sequence the use of tools for a given prompt
February 18, 2025 at 8:53 AM
It is AI though.
You use the AI LLM as an orchestration layer, then provide it with any number of tools (calculator, api access, code interpreter, search, etc).
The AI acts as an agent to recognize intent and sequence the use of tools for a given prompt
You use the AI LLM as an orchestration layer, then provide it with any number of tools (calculator, api access, code interpreter, search, etc).
The AI acts as an agent to recognize intent and sequence the use of tools for a given prompt
And yes you are right you usually wouldn’t use an LLM for simple arithmetic.
The value of the calculator comes when you ask it to do multi step reasoning that requires intermediate arithmetic step.
The value of the calculator comes when you ask it to do multi step reasoning that requires intermediate arithmetic step.
February 18, 2025 at 8:47 AM
And yes you are right you usually wouldn’t use an LLM for simple arithmetic.
The value of the calculator comes when you ask it to do multi step reasoning that requires intermediate arithmetic step.
The value of the calculator comes when you ask it to do multi step reasoning that requires intermediate arithmetic step.
The o series models use a combination of Internet search and chain of thought reasoning to get pretty good/factual results.
Of course it is still searching the internet so the output data will only be as good as the input data it parses.
Of course it is still searching the internet so the output data will only be as good as the input data it parses.
February 18, 2025 at 8:46 AM
The o series models use a combination of Internet search and chain of thought reasoning to get pretty good/factual results.
Of course it is still searching the internet so the output data will only be as good as the input data it parses.
Of course it is still searching the internet so the output data will only be as good as the input data it parses.
Newer versions delegate to a calculator tool for math so it should always work.
February 18, 2025 at 8:18 AM
Newer versions delegate to a calculator tool for math so it should always work.
But the older techniques can still be very efficient for certain applications
February 18, 2025 at 7:55 AM
But the older techniques can still be very efficient for certain applications