Some key points from the presentation include:
* The tool can prove mathematical theorems exponentially fast, even on a single core. * It uses a combination of numerators and primitives to prove theorems. * It has limitations, such as only recursing on row numbers and not being able to learn or predict the next token in a symbolic transform. * The tool will be useful for programming and solving big problems. * A playground will be released soon, allowing users to experiment with the tool and have fun using it.
Overall, the speaker is excited about the potential of the tool and believes it could be a valuable addition to the field of AI research and development.
01:53 Proving the Ruma Hypothesis - https://youtu.be/GddkKIhDE2c?t=113
05:13 Restricting Word Types to Exactly Two Bits Set - https://youtu.be/GddkKIhDE2c?t=313
05:18 First-Class Enumerations for Computation - https://youtu.be/GddkKIhDE2c?t=318
06:35 Enumerations Interoperability - https://youtu.be/GddkKIhDE2c?t=395
06:44 Narrowed List of Functions with Restrictions - https://youtu.be/GddkKIhDE2c?t=404
07:50 New Definition Emerges - https://youtu.be/GddkKIhDE2c?t=470
08:18 Section Title: Wrong Function Example - https://youtu.be/GddkKIhDE2c?t=498
08:24 Most Overfit Functions, But First One Stands Out - https://youtu.be/GddkKIhDE2c?t=504
09:40 General Addition Implementation Tested - https://youtu.be/GddkKIhDE2c?t=580
09:49 Finding Unique Functions - https://youtu.be/GddkKIhDE2c?t=589
10:36 Examining Functions with a Simple Loop - https://youtu.be/GddkKIhDE2c?t=636
10:38 First Argument Relationship - https://youtu.be/GddkKIhDE2c?t=638
10:48 PubShadow Literature Finds Problem in 0.06 Sec - https://youtu.be/GddkKIhDE2c?t=648
11:37 Complexity Hides in Simple Tasks - https://youtu.be/GddkKIhDE2c?t=697
12:11 Fast Loading Due to Under-the-Hood Optimizations - https://youtu.be/GddkKIhDE2c?t=731
13:33 Function Finder in 4.5 Sec - https://youtu.be/GddkKIhDE2c?t=813
13:41 Accelerating the Process - https://youtu.be/GddkKIhDE2c?t=821
14:37 Speedups via Cluster Parallelization - https://youtu.be/GddkKIhDE2c?t=877
14:45 Boolean, Natural Number, Words of Fixed Length, Lists, Funct - https://youtu.be/GddkKIhDE2c?t=885 16:23 Proving a Theorem by Defining an Empty Body - https://youtu.be/GddkKIhDE2c?t=983
16:30 Recipient Receives First Bit of List - https://youtu.be/GddkKIhDE2c?t=990
17:06 New Goal: Proving Equation for Pairs - https://youtu.be/GddkKIhDE2c?t=1026
17:40 Section Title: Boolean List Conversion - https://youtu.be/GddkKIhDE2c?t=1060
18:23 Constructing Equates in a Function - https://youtu.be/GddkKIhDE2c?t=1103
19:07 True Case Match: Convert Not True to True - https://youtu.be/GddkKIhDE2c?t=1147
19:21 Proof Complete: No More Obligations - https://youtu.be/GddkKIhDE2c?t=1161
19:31 Proving Self-Computation with Auxiliary Function - https://youtu.be/GddkKIhDE2c?t=1171
19:59 Proving List's Tail - https://youtu.be/GddkKIhDE2c?t=1199
20:39 Inductive Hypothesis Applied to Rest of List - https://youtu.be/GddkKIhDE2c?t=1239
21:29 Proof of Empty Pattern Match - https://youtu.be/GddkKIhDE2c?t=1289
21:32 Section Title: Synthesizer's Turn - https://youtu.be/GddkKIhDE2c?t=1292
22:01 Demonstrating a Key Theorem - https://youtu.be/GddkKIhDE2c?t=1321 23:05 Instant Search Proof - https://youtu.be/GddkKIhDE2c?t=1385
24:01 Enumerating the Proof Space - https://youtu.be/GddkKIhDE2c?t=144126:06 Integrating with Theorem Provers - https://youtu.be/GddkKIhDE2c?t=1566
27:20 Closing Notes: Simplistic Tool for Fast Solutions - https://youtu.be/GddkKIhDE2c?t=1640
28:08 Ready to Paralize with Cluster of Matchminis - https://youtu.be/GddkKIhDE2c?t=1688
28:18 Predicting Next Token - https://youtu.be/GddkKIhDE2c?t=1698
29:18 Predicting Next Tokens with Symbolic Transforms - https://youtu.be/GddkKIhDE2c?t=1758 29:33 Synthesizing Token Predictions from Lists - https://youtu.be/GddkKIhDE2c?t=1773
It's about the use of interaction nets, which gives an optimal evaluation strategy for the lambda calculus. I'm not an expert on it, but from my understanding it allows extensive sharing of computation across different instances of an enumerative search.
Parallelism of the computation is another big selling point, except modern hardware design is not well suited for the calculus. The author of the video recently tried to get the system to work well on GPUs and ran into issues with thread divergence. I think their current plan is to build some sort of cluster of Mac Minis due the good performance of the CPUs on that platform.
If this computation paradigm advances far enough and shows enough promise, I would expect to see companies start to prototype processors tailor made for interaction nets.
Would someone be able to do a tldr of what this is about? Looks interesting but the title doesn't give me a clue about what it is... Llm, audio,...?