This market involves two different questions.
Will SB 1047 become law?
Resolution criteria: This market will resolve as YES if:
It is signed into law by the Governor of California
Will Meta open source Llama 4 (or equivalent)?
Resolution criteria: This market will resolve as YES if:
Meta releases the full model weights and architecture of a large language model named "Llama 4" or releases a model that is a reasonable observer the successor to Llama 3 (regardless of its name).
AND the released model is made available under an open-source license (e.g., MIT, Apache, GPL)
This market will resolve as NO if:
Meta does not release a model meeting the above criteria by 2029
OR Meta releases the full model weights and architecture of a large language model named "Llama 4" or releases a model that is a reasonable observer the successor to Llama 3 (regardless of its name) and that model is not made available under an open source license.
Additional notes:
The release must include sufficient information and resources to allow independent researchers to run and fine-tune the model.
1,000
3.00@Bayesian I dunno, I'll have to look into it. I guess does the medium sized llama 4 seem a successor to llama 3?
@NathanpmYoung I guess I'd say no but it's really a fine line? basically my understanding is:
L3 was 405b, L4 maverick is 402B, so basically the same size, whereas L4 behemoth is 2T so bigger
L3 was trained on around 5e25 flops, L4 behemoth around 5e25 flops as well, L4 Maverick around an OOM fewer flops according to epoch estimate iirc
L4 maverick outperforms L3 405B, but L4 behemoth more so
L4 maverick seems to clearly be a successor to L3 in general (maybe it's the successor to 70B, as a medium size model for its time?), but if we're looking at L3's best, 405b, it seems like the successor is behemoth, and L4 maverick is similar in size and capability to 405B but is not their flagship, whereas L3 405B was
@NathanpmYoung That's currently unknown, but I'd be surprised if they didn't make it available in any way. it's unclear whether they'll open weight it like they did with Maverick and Scout
@CalebW With the binary way the question is framed, it seems quite weird to see 2 out of 2 released models being open sourced, and then saying "maybe/probably the third model released later will be open source, so we might resolve this question to 'Llama 4 is not open source'".
Judging from Chatbot Arena, even Maverick is clearly competitive with other current frontier models in a way expected for Llama 4. We've grown used to continuous releases spread over sometimes quite long time frames – but it does seem like this was the "main" Llama 4 launch, no?
Saying all this with a "bought it up to 95%"-bias, of course.
i don’t think Maverick is competitive with current frontier models, much less the future ones. Behemoth wil’ be competitive with current frontier models, not future ones. The lmarena score is ~fake because unlike other companies meta trained a custom maverick model to have a style that performs well on the lmarena. that’s worth a few dozen elo points. See eg