OpenAI’s math breakthrough played to AI’s strengths
- Thread starter JournalBot
- Start date
What source did you provide? All I see is that this was a general purpose model trained for this task. Many people believe that represents the best use of this technology: bespoke models specifically and exhaustively trained on curated data to perform well-considered and defined tasks.
The chain of thought document describes the model taking a largely brute-force approach, pulling up various existing theories that fit the problem and running through the exercise, in a manner that calls to mind Tsimerman's comment that his work on an approach like this was so tedious it "wasn't worth the effort." Lo and behold, here the model proves its true value: it does not value its time the same way humans do and will stick at mechanically plugging away until something seems to fit. According to the article and the links, with all the resources and tokens OpenAI can throw at it, it only found the answer half the time during its trials.
This is a worthy milestone in the development of these models; the downside is it will of course spawn more people to mistakenly conclude adding more monkeys to the group throwing things at the wall to see what sticks is "thinking." And heaven knows OpenAi can use the $500 prize for proving why Erdös himself struggled to prove his theory, even while falling short of solving the question itself.
edit: I see as I work my way through the comments MilanKraft also calls out the infinite monkeys analogy...
Last edited:
Upvote
0
(3
/
-3)
I was finding interesting the dude named after Lenin's confidence that he was among the enlightened and therefore suitable to cast judgment on the perspectives of others on how society's resources are best employed.
Upvote
1
(1
/
0)
I'll subscribe to this newsletter, with a note to the editor that it is less that I object to using these tools to do those things, it is misrepresenting what using those tools means AND that our choice to do so without consideration affects who we are.
I can:
1) Paint something, take a photograph, look at my pantry and employ my life experience to put what's there together for a tasty dinner;
2) Buy a picture or photograph, a paint-by-number set, use GIMP to manipulate and mask the errors in my photography skills, look up a recipe or buy one of those pre-assembled kits and cook dinner with it;
3) Commission a painting or photograph--leaving out as irrelevant to this point while certainly relevant to society--whether my patronization is of a human artist or a generative model, have them revise it to the taste I failed to express when originally requesting it because I didn't know what I really wanted, go to a restaurant and have someone cook for me, buy a package at the store and stick it in the microwave, etc...
All of these things are valid approaches to obtaining a product of creation. How we feel about which tier of approaching this is meaningful in an individual and collective sense.
Upvote
4
(5
/
-1)
Hey, it's Zen. Enlightenment will come to us when we embrace
Upvote
1
(1
/
0)
Acknowledged and glossed over even in detailed discussions with the major players, the plan has always been:
1) Maximize what can be achieved using current methods and all available resources;
2) [Discontinuity Breakthrough on the order of discovering Heechee technology]
3) Profit and World Joy for All!
Upvote
-2
(1
/
-3)
Think about that. "After it was announced" other models could get there if pointed in the right direction. Future models will have that knowledge in their training data. Incremental improvement is to be expected if not required just to keep pace--which is why training of models shouldn't be considered one-time start-up outlay. It is an ongoing (and opaque) component of operating cost, measured over lifecycles that are pretty darn short.
Upvote
-3
(0
/
-3)
Is this before or after we spend eight pages of comments regurgitating debates in other threads about what definition of "understanding" we start with before people who really want us to accept that "AI" is Just Like Real Folks move the goalposts on us?
Which some will do to avoid admitting the answer to your question, at the end of the day, is "the latter."
[ducks]
Upvote
0
(2
/
-2)
I asked what source YOU provided. Are you referring to saying "RTFA"? The article AND the links in it all say it's an unreleased general purpose model, but say nothing else, and I suggest you are likely over-stating how generic this model and its training may have been. Very few rational people are going to accept that this one time OpenAI is giving us all the information.
Upvote
-1
(2
/
-3)
Disproof and proof are different things. The article and the ones you link to explain this.
Upvote
0
(2
/
-2)
Dunno about you, but I had a lot of math and many of the questions involved filling in a blank of some sort.
edit: in no way am I suggesting that the filling of many blanks is non-trivial. I am saying, and have alluded to this in comments above this one, that the ability to frame the right question, and of whom to ask it, is often far more important. Yes, that acknowledges the value in so-called "prompt engineering" AND the value in knowing which knife to use to pare the apple and which knife to use to debone the trout.
Last edited:
Upvote
-2
(1
/
-3)
Somewhere in there is a reference to a table provided by OpenAI indicating it was a fifty/fifty process, which I read to mean on multiple trials of an unspecified n about half the time it found this answer, and half the time it didn't. Also unspecified is who did the initial validation before declaring Eureka! and sending this out to the mathematicians whose evaluations were subsequently part of their PR campaign. In the comments of those mathematicians were references to buzzes in that community over the model's apparent success before it went wide.
Upvote
-1
(2
/
-3)
I am going to stun people who want to put me in a box of their design and say something positive about these tools: They are excellent at identifying possible connections that represent a true synthesis, in the spirit of what Johansson called the Medici Effect.
What they have yet to show is that they have any ability to tell if the work they bring home adds true value, or if it's something only a parent would hang on their refrigerator.
Upvote
2
(3
/
-1)
You won't even look at what YOU wrote. You said you provided a source. Which source? I found none offered by you in this thread other than the "RTFA" comment. The "FA" does not go as far as to say what you assert, either.
Upvote
-1
(1
/
-2)
A "cope." Noted for future reference.
OpenAi has been vague on details here. What we can glean is that the process they used had an error rate of fifty percent. Using the same data, the same model, getting the right answer from a tool capable of producing it both before and after they knew they could do it was a coin flip. The target did not move. The tool was the same. Did the model they used even identify "This one works. This one does not"? We don't know. The overall process, which tends to be what I care more about, did make that determination, and this is a milestone. Absolutely this was a problem whose solution, for longer than my lifetime, was impractical to solve using other methods.
I applaud the developers of the tool and the mathematicians guiding it, validating it, and helping us understand what this means, and ask the reasonable question: "Can we know more about the true cost of this coin flip tool?"
I see this success in that context, rather from the context of "Eat it, anti-AI people!"
Sorry if that harshes your mellow.
Upvote
4
(4
/
0)
That's not at all what that article was saying. It was saying when provided with assumptions identified as false, the models were prone to using the false assumptions in forming their output without regard for their lack of validity.
Upvote
1
(3
/
-2)
There are two kinds of people in this world: Those that insist on putting everybody into boxes.
Agreed!
Upvote
1
(2
/
-1)
- Status