benbye - do you have any intuition why the model only needs 4 frames to work? I played with LeWorldModel and I am mystified why the latent prediction frame number is so small. It's an almost Markov like property where a short history is packaged as state.
I sometimes wonder how much of intelligence is being good with tools.
I feel pretty averagely smart but give me some good tooling like a good editor, a good type system, semantic grep, good testing and some solvers and I can actually deliver some work.
Maybe the trick isn't 500 billion parameters but a model super integrated with the task at hand for iteration and debugging?
FWIW the article really mirrors my own experience. I can run a small gemma4 for quick edits (and it's fast!) or data cleanup but for other tasks you do need a different tool (claude).
Another useful "Emergency exit" is CTRL+Z which stops the process and cannot be intercepted.
It's often faster than hitting CTRL+C and waiting for process cleanup, especially when many resources are used. Then you can do e.g. `kill -9 $(jobs -p)` to kill the stopped tasks.
Super useful tool but need to be aware that this is reading potentially untrusted input (e.g. in the case of http request logs) and written in c++, so a possible attack vector. I use lnav where I trust the logs, but do wish a safe implementation existed.
There is precedent: with type checkers like pyright you can opt into specific checks, or have a basic, standard, strict setting, each expanding the set of checks done.
How would dependencies work in this schism? E.g. if serde starts using named impls, do all dependencies have to use named impls?
I can't fully answer your question but I did once spend about a week porting plain internal configuration to cue, jsonnet, dhall and a few related tools (a few thousand lines).
I was initially most excited by cue but the novelty friction turned out to be too high. After presenting the various approaches the team agreed as well.
In the end we used jsonnet which turned out to be a safe choice. It's not been a source of bugs and no one complained so far that it's difficult to understand its behaviour.
I feel you (like many people) got burned by the steep learning curve. Empirically some pretty high powered companies use nix successfully. It's of course always difficult to know the counterfactual (would they have been fine with ubuntu) but the power to get SBOMs, patch a dependencies deep in the dependency stack, roll back entire server installs etc. really helps these people scale.
nixpkgs is also the largest and most up to date package set (followed by arch) so there's clearly something in the technology that allows a loosely organised group of people to scale to that level.
I've looked into this but saw hugely variable throughput, sometimes as little as 20 MB / second. Even if full throughput I think s3 single key performance maxes out at ~130 MB / second. How did you get these huge s3 blobs into lambda in a reasonable amount of time?
After reading it I found it much harder to enjoy movies showing bad security though (such has heists, nuclear anything, ..).
E.g. from the book I learned about the IAEA recommendations for safekeeping nuclear material [1], and it's pretty clear that smart people spent some time thinking about the various threats.
Anyway, rambling. It's a great and very entertaining book, go read it!
I think this is a common misunderstanding. The images are in the public domain. Nothing stops Getty (or you, or anyone) from selling them, even though you can just use them for free.
The value-add service that Getty offers is legal indemnification, i.e. they cover the legal costs if the image turns out to be copyrighted after all. To offer this service they spend some time and money upfront to research images' copyright status.
Whether you think that's good value for money is up to you.
More often than not recruiters (external and in-house) make up large numbers to get you to reply. Here is a verbatim quote from a mail I got last year:
> For the right candidate year 1 comp will be up to £500k.
After going through a three hour coding test, phone screen + onsite it turned out to be 120k + smallish bonus. Maybe I'm not the right candidate but most likely 500k was never on the table.
It's not the first time this has happened to me or people I know
either. What I'm getting at is that I'd take the Oxford Knight compensation numbers with a pinch of salt. Those 300k jobs _do_ exist but nowhere near as many as in the US.
Additionally levels.fyi or glassdoor data doesn't support a large number of 300k jobs in London.
The original SES doesn't seem to do anything to prevent meltdown/spectre attacks [1]
This version removed direct access "Date" [2] but I'm not sure I'd trust any code running in the same process space given how hard it is to fix spectre in general.
We're mostly using Protolude at the moment. It has some nice properties, e.g.
* "id" renamed to "identity" so you can use id in your own code
* panic functions like "notImplemented"
* a generic string converter "toS"
* lots more
But to be fair most modern preludes work OK.
Foundation is a bit different in that it doesn't ship a huge amount yet but has the potential to eventually replace the core prelude with saner default types like utf8-encoded strings. If you need to ship to production yesterday I would not use Foundation just yet.
I do agree but I also remeber that Haskell is 27 years old. Newer Haskell-inspired languages like PureScript don't have a built-in list any more.
There's a lot of old stuff in Haskell, e.g. String which is a list of char. We have a number of new preludes (base, foundation, protolude, ..) that improve the situation a lot, so I'm not sure we really need a "python 3" moment.
We could definitely be more aggressive in pointing out that you need to use a new prelude though.
I've been joking with friends that my next job will be AWS efficiency guru. I've somewhat optimized our own use, but I think I could use similar, simple rules to get 20% out of a 500k / month budget.
Give me what I save you in 2 months and I'll have a good business :)