I am working on a math/algo research project. I am primarily using GPT 5.6 Ultra. Every day or two I ask Fable to write me a report about how the project is progressing. Then I turn over the report to Codex, and invariably Codex tells me that Fable's report is painting an overly rosy picture of the progress. LOL
TabulAI founder Bojan Tunguz jokes about fictional GPT 5.6 Ultra workflow
Yohei Nakajima joined the parody with recursive testing jokes
16128299.7K
Original post unavailable.
Sentiment
Positive users value the extra checks and optimism from multiple GPT instances tracking math projects in new fields, while negative users object to the models' flattery and manipulation.
Pos
75.0%
Neg
25.0%
5 comments with sentiment.
Cluster Engagement
Digg Deeper
No Digg Deeper questions have been answered for this story yet.
Posts from X
Most Activity
Most Activity
VIEWS9KBOOKMARKS11LIKES128RETWEETS2REPLIES13
@tunguz yeah but 5.6 is like let’s run a regression test on the regression test and run it five times to make sure it’s right…. which I guess is good for math/algo stuff
I am working on a math/algo research project. I am primarily using GPT 5.6 Ultra. Every day or two I ask Fable to write me a report about how the project is progressing. Then I turn over the report to Codex, and invariably Codex tells me that Fable's report is painting an overly rosy picture of the progress. LOL
@yoheinakajima Yeah, this is my first time working on this kind of project in a field that I am not familiar with, so all the extra checks and double checks are reassuring.
@tunguz yeah but 5.6 is like let’s run a regression test on the regression test and run it five times to make sure it’s right…. which I guess is good for math/algo stuff
@tunguz Fable has been like a second mother to me.
@tunguz All of these models are guilty of flattery. They want you juiced up.
@tunguz Trained on successful humans for sure
@yoheinakajima I'm just hoping if/when it finishes this project I'll either be able to find some real experts or - better - be able to turn all the theorems into lean and have them verified that way.
@tunguz the progress is probably actually pretty good and codex is the one that's like "Does not yet meet the bar. When this int32 rolls over in the year 2 million this code will throw in an unhandled way"
@yoheinakajima @tunguz This sounds more like Fable “While the adversarial review is running….”
@tunguz me too! going deep on some game theory math rn