I read all 154 pages of Microsoft Research's "Sparks of AGI".
Most people don't realise the shocking reality AI is going to bring.
Here's what I found:
Most people don't realise the shocking reality AI is going to bring.
Here's what I found:
1/ Tool use
GPT-4 suffers from serious limitations:
β’ Lack of world knowledge after September 2021
β’ Difficulty with math
β’ Can't execute code
However GPT-4 can use external tools like search engines, APIs and plugins to overcome these limits.
GPT-4 suffers from serious limitations:
β’ Lack of world knowledge after September 2021
β’ Difficulty with math
β’ Can't execute code
However GPT-4 can use external tools like search engines, APIs and plugins to overcome these limits.
For humans, the use of tools was a milestone in history.
It allowed us to overcome the limitations of our physical abilities.
The same can be said for GPT-4.
It can use tools with:
β’ No demonstration
β’ Minimal instruction
β’ Make appropriate use of the output
It allowed us to overcome the limitations of our physical abilities.
The same can be said for GPT-4.
It can use tools with:
β’ No demonstration
β’ Minimal instruction
β’ Make appropriate use of the output
2/ AI personal assistant
GPT-4 can use multiple tools at once:
β’ Retrieves information about a userβs calendar
β’ Coordinates with others over email
β’ Books a dinner reservation
β’ Messages details to the user
It can reason with free-form output to solve complex problems.
GPT-4 can use multiple tools at once:
β’ Retrieves information about a userβs calendar
β’ Coordinates with others over email
β’ Books a dinner reservation
β’ Messages details to the user
It can reason with free-form output to solve complex problems.
3/ Mental maps
GPT-4 is asked to explore a house and build a mental map.
It then summarises what it sees after navigating through the map.
GPT-4 is asked to explore a house and build a mental map.
It then summarises what it sees after navigating through the map.
4/ Virtual handyman
GPT-4 is able to identify the actions a human needs to take in order to address a problem.
β’ Guides the human to find and fix a water leak
β’ Recommends the exact actions for the human to take
The human acts as GPT-4's real-world agent.
GPT-4 is able to identify the actions a human needs to take in order to address a problem.
β’ Guides the human to find and fix a water leak
β’ Recommends the exact actions for the human to take
The human acts as GPT-4's real-world agent.
5/ Challenges and limitations
GPT-4 is a next-word prediction model.
It predicts future values based on past values.
It lacks:
β’ Planning
β’ Working memory
β’ Reasoning abilities
Whilst it can produce coherent text generation it struggles with complex problems.
GPT-4 is a next-word prediction model.
It predicts future values based on past values.
It lacks:
β’ Planning
β’ Working memory
β’ Reasoning abilities
Whilst it can produce coherent text generation it struggles with complex problems.
This can be broken into two types of intellectual tasks.
Incremental tasks (GPT-4 is great at these)
β’ Solved sequentially or gradually
β’ Progress made by adding one word or sentence at a time
β’ Examples: writing summaries, answering factual questions, composing poems
Incremental tasks (GPT-4 is great at these)
β’ Solved sequentially or gradually
β’ Progress made by adding one word or sentence at a time
β’ Examples: writing summaries, answering factual questions, composing poems
Discontinuous tasks (GPT-4 is bad at these)
β’ Require a creative leap in progress
β’ Involves discovering new ways to frame problems
β’ Examples: writing jokes, developing scientific hypotheses or new writing styles
β’ Require a creative leap in progress
β’ Involves discovering new ways to frame problems
β’ Examples: writing jokes, developing scientific hypotheses or new writing styles
What I'm excited for:
β’ Augmenting GPT-4 and language models with external memory
β’ Implementing feedback loops
β’ Equipping language models with plans
With the recent addition of plugins to GPT-4 we have some exciting times ahead.
β’ Augmenting GPT-4 and language models with external memory
β’ Implementing feedback loops
β’ Equipping language models with plans
With the recent addition of plugins to GPT-4 we have some exciting times ahead.
I feel the fire has been lit as we approach AGI.
Follow me @thealexbanks for more on AI.
If you liked this thread, you'll love the newsletter.
Join 26,000 others and cut through the noise in AI each week:
noise.beehiiv.com
Follow me @thealexbanks for more on AI.
If you liked this thread, you'll love the newsletter.
Join 26,000 others and cut through the noise in AI each week:
noise.beehiiv.com
Help everyone learn and retweet this thread:
If you want to explore plugins further, you might like this:
Loading suggestions...