I hadn't seen SwiReasoning (https://swireasoning.github.io, paper and code), it looks like that works at generation time without any requirements on the model. It increases token-efficiency and accuracy, but at first skim it seems like this would be incompatible with multi-token prediction. For large reductions in token budget it could be worth it.
As I understand it the basic premise of all the speculative decoding schemes is that the logits on the draft don't need to be exact so long as you mostly sample the same tokens, and because each position is fed by the embedding associated with the previous position's token you sort of "round away" error. With SwiReasoning I think you skip the sampling/rounding part and do something continuous using the whole distribution, so it would seem to rely on the accuracy of those values. MTP still makes sense outside the latent reasoning chunks though.
"The Middle-east" isn't a war zone. Even the parts of the middle-east that are, don't have any drone deployments. Lebanon maybe? Reports are thin.
Maxar is/was primarily a satellite data company, and to say Pokemon data would add any major value in any of today's active drone deployments with the level of Satellite coverage Maxar already has is a wide stretch.
Moreover, ground forces in the area would need pretty heavy jamming tech in place too for this kind of data to be useful. It's a sliver of a sliver of a sliver situation.
I am a daily player, I have scanned something once, the rewards were minuscule, I never did it again. I have that specific vivillon which was hard to get because not many players were from the relevant area even before the current events, and I just can't see how the war is related to any of this.
Delaying a technology release is not going to stop that in the long term. Society, culture, and the support tooling just needs to adapt. Just like how AI coding is still in the early days.
The sooner people learn the risks and build the infrastructure to make it fail less the better.
If I'm reading that right this covers all of Meta's state lobbying spending which is $45M + $25M federally and it's not all directly related to this age gating stuff, but general shaping anti-child exploitation policy?
> On every social media regulation bill in Colorado, Meta takes an "Amending" position, actively fighting changes. Across 117 lobbying records on 22 bills:
> Bills regulating social media: Meta position is "Amending" (fighting)
> The one bill putting the burden on OS providers: Meta position is "Monitoring" (watching)
> Meta fights bills that regulate Meta. Meta watches bills that regulate everyone else.
So their lobbyist choosing not to fight against the OS age gating is the big reveal.
There was an AI generated slop report posted to Reddit that went viral. It even got some news coverage before anyone stopped to read the files and discover that it was gibberish.
I still get downvoted for pointing it out and trying to ground the conversation in facts. As you noticed the story continues to thrive on bad news sites and social media.
Meta has not spent $2 billion lobbying for this or on lobbying altogether.
It’s amazing how that one AI slop project that made this claim got so many people to believe this number.
Spreading this disproven AI slop around isn’t helping. It just makes opposition look like uninformed conspiracy theorists who can’t fact check anything.
Meta is lobbying to get age verification installed at the OS level so that they don't have to bear the cost themselves. I know nothing of the $2B but the fact that Meta influences the legislation can hardly be denied:
During the meeting, Meta executives, including Antigone Davis, global head of safety, are understood to have argued that any age checks should be handled on a smartphone operating system rather than by the likes of Meta.
reply