$28m actually seems worse. If I wake to a $100b bill, that’s obviously a mistake. If I wake up to a bill in the millions then my first thought would be “oh no what did I do wrong, this will ruin my life”
The current taalas chip is for a 3.1B param model. I’m hope so much that they can get that up to the 30B range. Just imagine Gemma 4 or Qwen 3.6 at 17k tps.
I believe they got rid of the smaller battery. Before the tax credit was removed, they were sourcing a more expensive battery. After the credit was removed the switched to a cheaper battery. The cheaper battery would not have qualified for the tax credit, but that’s not an issue anymore.
I was browsing Amazon and I found one “author” with over 100 published in the last 12 months. And they weren’t cheap either. Most were $50 or more for the paperback version.
It really put a sour taste in my mouth. At that volume (2 books per week) you’d think Amazon could identify and label that it’s AI generated.
> Under this commercial agreement, Meta will provide funding to support the deployment of the Natrium plants, with delivery of initial units as early as 2032
The wording there implies some upfront money from Meta, and that this isn’t just a PPA like we normally see.
But with no numbers attached it’s hard to know if it’s a serious investment or just PR fluff.
> Unlike our interventions for cybersecurity, biology and chemistry, and distillation attempts, these safeguards will not be visible to the user. Fable 5 will not fall back to a different model. Instead, the safeguards will limit effectiveness through …
No it won’t fall back to Opus, they will purposely return dumbed down or tainted information with the goal of the end user not knowing the results have been impacted.
For sure not everyone can self host (though I suspect most tech and tech adjacent companies can).
But all the existing “safe/trusted” cloud providers will offer cheaper models. The choice won’t just be GitHub vs OpenRouter. It will be DeepSeek, qwen, Gemma on GitHub vs Opus on GitHub. I’m sure AWS, GCP, Azure, etc will be happy to sell open weight models. And those lower priced models will put a cap on how many people pay for higher priced models.
> but most of them will just pay 10 or 20 times or more for AI than they used to pay until now.
At my work, after the GitHub price hike, we all got the option of 1. Keep using Github, but with the same total spend as before. i.e, use 1/20th as much since the dollar cap isn't changing. Or option 2, use as much self hosted DeepSeek v4 Pro, Qwen 3.6, Gemma4 as you want since it's almost free. (to be fair, we already had the GPUs)
If the Chinese keep releasing open weight models that are "close enough" to the big 3, I expect many orgs to make the same choice. I think we will see VPs and higher start saying "you better have a good reason for using Opus 4.8 or GPT 5.5 Pro, do you have metrics showing ${cheaper_model} isn't good enough?"
I just did an install to add solar and batteries to my shed to power lights and an AC. It was pretty easy. Hardest part was flattening the ground since I did a ground mount system. 5kw panels and 5kwh of batteries. $1000 for the panels, and $1,400 for the battery and inverter. $250 for the ground mount. Plus a bunch of miscellaneous expenses (tools, wires, permits, etc). It would be cheaper if I did it again since batteries and inverters seem to get cheaper every 6 months.
The best part of a quiz everyday is that in addition to the testing effect, you can easily fit in the spacing effect and interleaving effect. It’s a rock solid combo, that is well studied. We have pretty strong evidence that it works for all students in all domains.
Apparently there is a way to solve this without brute forcing all the combinations. It has to do with looking at how many even an odd numbers there are, and taking into account the goal number is odd. And then thinking through the combinations [even-even=even, even-odd=odd,…]
Though this is obviously not something I would expect a 4th grader to solve.
In seriousness, Go is a good choice. Or at least it’s not a bad choice, I’d definitely pick go over many other languages. If go had a better type system, it would be damn near perfect.
I tried zed sometime ago, and the limiting factor was devcontainer support. It looks like they’ve made some progress there https://zed.dev/docs/dev-containers
I’ve never worked with SmallTalk professionally, so I could be mistaken. I know you can write/compile SmallTalk source code. But my understanding is that this wasn’t the way it’s intended to be used. I am under the impression that the live images were how the language designers intended it to be used. And that the live images are a better representation of the OOP/message passing paradigm.
I’ve very slowly been trying to do the “99 problems” list in each of these languages groups. It’s been a fun experience seeing the differences. Though I think I would need a larger, less algorithmic, project to really see each group’s strengths. Especially for the OOP group.
One thing the article didn’t touch on was SmallTalk’s live visual environment. It’s not a normal source code / text language.
Is that available in the US? Can you share a link? That’s an amazing deal. I’ve been recommending server rack batteries (5kwh for $750) to people but if there is something better I’d love to see it.