digdugdirk

Do you have a link to that statically typed Fennel compiler? That sounds fascinating.

I know this sounds ridiculous, but try fine tuning a small local shell -> plain language model, then using the output of that for whatever you need on your end. You'll get the gist of what the frontier model is trying to accomplish, and can handle it with whatever harness workflow you want.

Pinned at the front and back on a rotating axis down the centreline of the car, with the "car" itself being mostly lightweight composite with zero internal components other than the necessary air tanks and a large RC car motor to get it to roll up to the jump point? For the support structure, just put some dark matte scaffolding in the background so it stays out of shot and hidden by the light show. It wouldn't be too far off from standard wirework for movie stunts, and the compressed air could still do enough work to fool the eye. It's not like people have an innate sense for what a gas thruster powered flying car should look like anyway.

But you know, that's just off the top of my head.

They were aiming for better concurrency, I believe. If you're rewriting something regardless, might as well use a more modern language that's easier to reason about during development.

Did you read the article? Meta themselves aren't confident that what they're doing is legal, and they're very clear about the fact that it might be clawed back by the IRS because they're abusing a tax credit intended to be used for experimental research by putting their massive data centre build out into that "experimental research" bucket.

If a company's accountants and lawyers are concerned enough to bury disclosures in their securities filings to cover themselves, but the company leadership is brazen enough to proceed anyway, I'd say that's plenty of justification for people to be mad at the executive leadership.

The US Petroleum Reserve is governmental, not done by the oil industry. The US government buys and stores oil when prices are low to provide a buffer for volatility in the markets or for things like natural disasters. The reserve being low means that the oil companies are very close to having a market that has no alternative but to pay whatever price is asked for oil.

To your point about oil and LNG being expensive to transport - that just means oil corps can turn down production at the locations impacted, and jack up the prices for whatever is produced in the locations that aren't. Lower operating costs + higher revenues = profits.

It really is a perfect storm for massive returns for the major players. As a bonus, it's impacting the Saudi and Russian national players, which are their toughest competition since they don't need to file any quarterly reports. I agree that it can't go on forever, but the oil industry is rolling around in money like Scrooge McDuck in the meantime.

Right, that's exactly my point though. Those are absolutely valuable use cases. But are they useful enough to justify a trillion dollar valuation? Or is most of the economic value in the stuff that already exists, and can be performed nearly as well by qwen/deepseek/kimi/GLM/etc?

And beyond that - how long until those individual Astra capabilities are distilled into separate Qwen-27b size models, with harnesses and scaffolds specifically designed to support that functionality?

The difference now is that they've hit the "good enough" point. LLMs are a tool, and that tool is useful but not incredibly valuable unto itself.

To make a manufacturing analogy - ChatGPT was a manual machining mill, and in the years after we've gone from that to a 3-axis CNC mill. Now we've added a 4th and 5th axis, which is great for the 2% of parts that need that functionality. But the big win was that initial jump from manual control to CNC. Why would I pay an extra $2 million for my CNC machine when I could just design my parts to be simpler to produce instead? The AI labs are trying to make these incredibly complex tools, but the market doesn't want/need them so they're competing on price for the tools that people do use. By selling their metaphorical CNC machines for half of what they cost to produce.

Oh, and we've bet the entire economy on the hope that fancier CNC machines will magically solve all our problems in all industries, from healthcare to the legal system.

So - will AI progress continue to improve? Sure. Will we continue lighting money on fire in order to make it happen? That remains to be seen.

Thanks for posting this! Love your team's work.

Ha. Amazon's in-app AI agent is called Rufus. It's incredibly annoying, and they love directing you to an 'Ask Rufus' page where you watch an AI agent not-quite-finish repeating your search terms back at you before you realize you're not actually on your search results page, and then spend some additional time trying to find the link to take you to said search results page.

I assume somewhere there's a team hitting their AI query KPI's due to this, but I really wish they'd shut off a few server racks and save everyone the annoyance instead.

I've seen previous needle releases running on esp32s - any idea how well this new one would run on something like the newer esp32-p4?

Of course it was. They clearly decided that the benefit to the company valuation was higher than the potential downsides when announcing to the world that they committed a criminal act via negligence.

If it wasn't a marketing stunt, they would have at most quietly settled any legal matters with huggingface behind the scenes, fixed their evaluation harness so it wouldn't happen again, and avoided the potential future liability.

It sounds like your issue is with capitalism, not with renewable energy. Luckily, we have plenty of examples of aspects of society that have been deemed too important to leave purely to market forces. In fact, electricity generation is already (mostly) one of them.

As solar and batteries become even cheaper (we're already at the point where they're cheaper than anything else) we can just choose to overbuild capacity and turn down/off whatever other generation isn't necessary at any given time. Utilities already have to predict grid usage based on weather and all sorts of other factors, so it's not like this isn't something that isn't already happening.

Not quite. Looped models do the extra "thinking" inside the model's layers. So the token gets twice the number crunching performed on it before it gets spit out. I think of it as the first loop "kickstarts" the process, and the second loop refines it.

I mean, that's cool and all, but the numbers are really going to shift when it accidentally goes off and orders that same hardware from every vendor in your local region and the top 5 online results for comparison.

It's the same problem as all other LLM solutions (that I hope OpenAI is working on!) it's non-deterministic, and there's no way for the user (or model provider) to know what the distribution of possible outcomes is. This just gets compounded when multi-call harnesses come onto play.