Mini-AGI
AI
continual learning model
SHOW HN2026
How HN took it
11 positive, 10 neutral or asking, 1 mixed, 18 critical, out of 40 comments about the project. Thread score -0.03.
Idea: 7 for, 6 againstPerformance: 3 for, 8 againstDocs: 0 for, 3 againstDesign: 1 for, 0 againstExecution: 0 for, 1 against
The thread, comment by comment
Each row is a top-level comment and its replies, in HN's order. Select a cell to open that comment on HN.
positive neutral or mixed critical about another commenter or something else the maker
What HN said
What people liked
Continual learning approach on a single stream is interesting
From 2 comments
Seems interesting, I've been messing with a lot of continuous learning approaches lately and it's cool to see something that's built from the ground up for avoiding catastrophic…
advael on HNI built a smaller scale replication. Most of your claims prove out but the lever is essentially Biderman et als "learn less forget less". My first pass actually made this mistake…
jmatthews on HN
Expert swapping architecture is nice
From 1 comment
The expert swapping architecture is very nice. Have you considered doing nested reinforcement learning where you use the nesting as a sort of low pass filter? The concept is as…
imtringued on HN
Objections
Claims of AGI and learning are overstated; outputs are not coherent
From 3 comments
If you actually scroll through the transcript he links to, you will see that something that looks like it could be training is happening, but no coherent responses are coming out…
ilaksh on HNI would be okay if you shared it as a potential idea and possibly interesting early result, but the language you are actually using to characterize it is misleading or…
ilaksh on HNTo pick a couple of examples: ``` --- chess --- prompt: '<g>1700 1-0 1. e4 e5 2. ' [raw] repeated 8-grams 83% Nxd5 Nxd5 Nxd5 Nxd5 Nxd5 Nxd5 Nxd5 Nxd5 Nxd5 Nxd5 Nxd5 Nxd5 Rxd5…
seanhunter on HN
No benchmarks or performance analysis
From 2 comments
Using the term AGI and not including any performance analysis. My AI calls it: "massive marketing overreach". Somebody called this slop in the comments. As a professor who…
synctext on HNThis is slop. 8M parameter dense model with context length 64 that you train on enwik9 in 2h will have 1.15 bpb. This model has 1.8 (bits per byte, lower is better).
lostmsu on HN
Name is misleading
From 1 comment
Mini-AGI is a totally inappropriate name - it seems what this project is shooting for, but not delivering on, is being a language model with "continual learning". Where it seems…
HarHarVeryFunny on HN
Requests
Run standard benchmarks and publish weights
From 2 comments
For sure. As it will pass through the whole corpus I will share the weights, run it through established benchmarks for small models and share all of this as an update. I am also…
volotat on HNI got you, will add later to the repository.
volotat on HN
Explain how catastrophic forgetting is avoided in the docs
From 2 comments
I read your github page and there's mention of continual learning but no argument how catastrophic forgetting is avoided. I think you should explain that better.
tnspacetime on HNI wish I could understand what a single graph in that nice graphic of graphs meant. No explanation for any vertical or horizontal axis. Looks pretty though.
Schlagbohrer on HN
Questions
Does the architecture generalize or only memorize?
From 2 comments
Is this architecture actually able to generalize or is it mostly based on memorization? Have you tried some basic tasks that require generalization? e.g. number addition etc?
cpldcpu on HNWhat's the advantage of doing this, versus becoming good at context management and RAG? I always found trained knowledge unreliable, given that it is lossy by construction.
awfm9 on HN
How does it behave after topic shifts?
From 1 comment
Have you tested what it remembers from early in the stream after a shift in the topics thrown at it?
ilusion on HN
Alternatives named
Standard LLMs and RAG
From 1 comment
What's the advantage of doing this, versus becoming good at context management and RAG? I always found trained knowledge unreliable, given that it is lossy by construction.
awfm9 on HN