atlex2

Actually, no - this is what controllability is all about. Humans have to understand it.

Would love to hear your thoughts on NVIDIA NeMo Switchyard...

Turns out those seams _were_ load bearing.

What was the architecture of your router? If it was based on GRPO/RL, it would be interesting to hear why your router performance capped.

I think the truth is that it's not an efficient cost cutting method. Your router has to be at least as 'smart' as all the but the smartest of your models (models do poorly when asked 'is this a task you're well suited to'), and that means you're caching multiple prompt histories including kv-filling/prefix caching on your expensive router model. Most of the time, not super great for savings.

I’ve started to pick up on some of the “unwilling to dig deeply into the humans perspective” & “provide ideation and then run with it” in 4.7. I actually think it’s consistent with confabulation, now that they’ve removed most of the models ability to observe its own reasoning in 4.7.

The effect is over-complicated engineering that takes way more time to review as to its right-size for the job.

Feels like hiding things, however.

A couple drawbacks so far via our scenario-based tests:

1. You can't ask the model to "think hard" about something anymore - model decides 2. Reasoning traces are no longer true to the thinking – vs opus 4.6, they really are summaries now 3. Reasoning is no longer consciously visible to the agent

They claim the personality is less warm, but I haven't experienced that yet with the prompts we have – seems just as warm, just disconnected from its own thought processes. Would be great for our application if they could improve on the above!

It probably still took way more time to write than it did to read.

It's also kind-of their point that they find the information delivery more important than the prose; they're leaning into their situation :-D

In January and February we flew two Air Tours with our AI copilot Shirley, here's what we learned about ASR, connectivity and models in-flight!

I built some of the first apps on the App Store. Top twenty navigation app. Won an ADA.

Still, my pocket is my iPhone pocket.

Until they release something the size of the X or smaller, I’m sticking with my iPhone 13 Mini or eventually going for a Razr style Android.

Every year they release something, I go check it out. My love for Apple dies a bit more.

I think spatial tokens could help, but they're not really necessary. Lots of physics/physical tasks can be solved with pencil and paper.

On the other hand, it's amazing that a 512x512 image can be represented by 85 tokens (as in OAI's API), or 263 tokens per second for video (with Gemini). It's as if the memory vs compute tradeoff has morphed into a memory vs embedding question.

This dichotomy reminds me of the "Apple Rotators - can you rotate the Apple in your head" question. The spatial embeddings will likely solve dynamics questions a lot more intuitively (ie, without extended thinking).

We're also working on this space at FlyShirley - training pilots to fly then training Shirley to fly - where we benefit from established simulation tools. Looking forward to trying Fei Fei's models!

Yeah - Tesla is in a weird level 2-4 space. They've managed to shunt liability onto their customers until now..

On the other hand, test-time weight updates would make model interpretability much harder.

They're building on a false premise that human equivalent performance using cameras is acceptable. That's the whole point of AI - when you can think really fast, the world is really slow. You simulate things. Even with lifetimes of data, the cars still will fail in visual scenarios where error bars on ground truth shoot through the roof. Elon seems to believe his cars will fail in similar ways to humans because they use cameras. False premise. As Waymo scales, human just isn't good enough, except for humans.