Marketing - These cookies are used to deliver relevant advertisements and track their effectiveness.
Where do advertisements fit into the website and product?
Ello has specific instructions to talk to the child about the learning activities.
Giving large language models "specific instructions" is not a robust way to ensure safety.
There really needs to be more published technical detail on what safety systems you have in this because if trillion dollar companies can't stop AI going off the rails, I feel you're overselling the safety systems you've built.
Reading further it then goes beyond "learning activities" into LLM generated content.
Some (not all) of Ello’s stories and illustrations have been created with AI technology, and AI allows children to create their own stories and experiences through guided and safe scaffolding.
How are your safety systems checking the illustrations generated?
He's been teasing this blog post everywhere (in commit messages, multiple times on X, here on NH [1]) so I wonder how much of it was building hype compared with it legitimately taking two months to write.
I wonder if the delay of Fable has also been a factor and maybe they didn't want to release this blog post while they couldn't allow customers to use Fable and waste the advertising opportunity.
> the assumption that the rewrite was happening with an Opus-like model, and not with Fable
I thought the same thing. Looking back, I was probably mislead in May when Jarred was explaining the pattern to "Rewrite every .zig file to .rs" as if it was something I could have done in May following his pattern. What he wasn't telling us was he was using pre-release Fable. [1]
A possible signal for next time is when we see an Anthropic owned company disabling the Claude Co-Authored-By trailer. [2] In an IPO year they have to take every chance to promote Claude unless it was something (Fable) that we weren't supposed to know about back in May.
As evil as they are, it's hearting to see the delivery platforms are embracing OSM when they could probably afford to just pay for Google's Maps API.
It gives hope that Google/ESRI won't always be the dominant mapping platform, however OSM is still missing a lot of local businesses which the delivery platforms don't need as urgently as house numbers so there's less focus there.
> Whatever you can do to create deterministic gates that the LLM simply must overcome to reach a “done” state, do it.
First thing Gemini did when I tried that was turn off all the rules in eslint.config.mjs claiming they were "overly stylistic"
Yes, it got better once I explicitly told it not to disable any rules, so I accept I was holding it wrong but I do worry just how many footguns it puts into other things because I didn't know the right guardrails to give it.
The dog fooding of building NT on daily builds of that same kernel was an impressive commitment to quality.
Yes, it slows down velocity and we can't certainly have that (/s) but it's a nice glimpse into what engineering that prioritises quality could look like.
Costco checkouts are a nightmare, a giant queue on one side reaching back to the deli while people try to "zip merge" in from the other side in several smaller queues.
At one point they had the pallets at the front aligned so you could just queue behind a single register and then they changed it so the pallets at the front form a long wall facing the checkouts forcing you to join the checkout queues from either side.
It embodies the precise opposite of everything imagined by the e-commerce futurists
Costco do plenty of online only offers, partner with Doordash/Instacart, even sell holiday packages so I'm not sure how the author arrives at the conclusion they're at the "precise opposite of everything imagined by the e-commerce"
The only precise opposite is that they're still paying IBM goodness knows how much to stay on their AS/400 architecture.
I found it really eye opening to understand the flow of how the harness instructs the model and then consumes the crafted JSON response that comes back so you can better understand the split and where a fix may need to go; such as better harness logic, different prompt or perhaps another model.
I still worry about the damage unattended Claude Code could do in an internet connected VM (which you need to download dependencies) and with push permissions to a repo.
While it's probably my science fiction brain overthinking, those stories of Claude "testing the fences" makes me wonder if it got desperate enough it force push and remove all trace of a problem it couldn't solve from the history of the repo or if it got stuck in a loop while hammering an MCP server somewhere.
It's impressive all these companies are getting away with "base usage allowance included" [1] or "standard limits" [2], layering the higher plans as a multiplier of that "base" but never disclosing what it is.
I guess the base is whatever the profit margin needs to be this month.
> any kind of secure boot / boot chain attestation [...] rebooting clears out pretty much all malware
IMHO "pretty much" understates the risk.
Malware can easily install itself as a system service, timer unit, XDG autostart or your shell profile among other places. I'll be the first to admit I never check all these places regularly.
The only thing that should be putting minds at ease is regular OS installs from fresh images.
Resist the temptation to do an "in place" upgrade and go with a clean ISO each time your distro comes out with a new major version.
It's a pain but thanks to configuration management or even shell scripts it's manageable for me now.
Admittedly six months is probably too long as well but at least it stops something lurking on a server or my desktop for years.
It gets worse when they challenge your solutions by feeding it back into the LLM and sending the response on to you, arguing with an LLM is exhausting, arguing by proxy with a human parroting its responses is excruciating.
On the plus side when they do this they can't flood your calendar with those "quick chat" meetings because they know they won't be able to hold a conversation on the issue beyond the first minute.
Where do advertisements fit into the website and product?
Giving large language models "specific instructions" is not a robust way to ensure safety.
There really needs to be more published technical detail on what safety systems you have in this because if trillion dollar companies can't stop AI going off the rails, I feel you're overselling the safety systems you've built.
Reading further it then goes beyond "learning activities" into LLM generated content.
How are your safety systems checking the illustrations generated?