crimsoneer

I'm so glad I learnt it before LLMs. I can't imaging trying to get good at Pandas/Polars coding nowdays when so much of my analysis is "Fable, please do this for me, k thx bye"

This was true a year ago, but definitely not today. Their biggest models are pretty much at the frontier, and Glimmer is arguably the best non-Chinese model in the "small" weight class (see my hacky benchmark http://botbowlbench.com/)

I think the problem is up to date knowledge isn't really the difficult or useful bit. We're sort of moving to a world where the real value in training comes from tool calling and performance over long context - your models shouldn't really be relying on their internal knowledge, but instead calling various tools. But the performance gap between now and six months for that is pretty huge.

Oh, I'd missed this, that's a shame! But yeah, Inkling + Reflection are properly open and quite good.

I added this to my weird little personal benchmark ( https://botbowlbench.com/) a couple of weeks ago, and it's...fine, but certainly not up to the best Chinese or US open weight models.

Inkling, Nemotron Super, Beam Reflection (and of course Kimi, Qwen, etc)

I think this is a quite optimistic view... releasing a SOTA model is a collosal capital investment, and unlike free software, it's not iterative - you can't make a small model and improve it one step at a time, you need to risk the whole load of cash, expertise and GPU time up front.

It's only happening now because labs have bucket tons of VC cash to burn and need a big public flex to raise more cash, or because they're subsidised by the Chinese government. I'm not massively confident either of those will stay true forever.

It is expensive as hell. That said, on my weird personal benchmark (https://botbowlbench.com/) it reliably beats Opus 5.5, which surprised me a fair bit.

This feels like a pretty big deal. Quite a few people sleeping on argon because it doesn't come via a subscription, but it's noticeably better than pretty much everything in my own weird game based benchmark (https://botbowlbench.com/)

> it's already known that their model is very likely trained on input gathered from scientists which use OpenAI as a tool for their research

Is this actually known or just broadly suspected? I could never quite figure it out...

This is just straight up, verifiably not true.

For the same reason every single open weight models gets released: it's a big flex and gets investors and developers excited.

Is it a sustainable model? Not really. Will it all blow up in our faces eventually? Maybe. But frankly, releasing code like this is far, far more sustainable than releasing whole models.

Honestly, I hate to tell you this, but the Germans have a whole track record of this fascism malarkey. I'm sure Vlad is pleased as punch, and definitely nudging it along, but if the influence was quite so impactive it would also be a lot more visible.

It's releasing open source code. This is a good thing for the world, even if it impacts the portfolio of US VCs.

But when is another open weights release coming out...

This is how science is meant to work.

Woah, this seems like a big deal (assuming the benchmarks are as good as claimed)?

Mistral slightly proving me wrong (and I'm not mad).

You don't get claim to be open and then not release your actual product.

He really likes skiing and being left alone.

The UK Team build GOV.UK Chat did a fair bit of blogging on how they managed this

https://insidegovuk.blog.gov.uk/2026/03/16/5-things-we-learn...

Worth recognising this was (and still is) mostly nonsense. There were loads of figures like "3 CCTV cameras per person" that included your local shop having a camera in the back, which has nothing to do with state surveillance. The police in the UK have significantly less power to surveil people than police in the UK. Council CCTV is shockingly rare now days.

If they'd wanted the data, they could have submitted an FOI and asked for it, but the journalists far preferred to just run "30 people a day arrested for hurty words" rather than "this is quite complicated but potentially problematic and we should dig into it"

Sigh, this is an absolute bullshit figure and people need to stop parroting it - it's 30 arrests under the mal comms act which might include for example, phoning your ex girfriend to say you're coming to stab her. Or phoning in a bomb threat via telephone. Or any number of pretty terrible things it's perfectly reasonable to get nicked for. Or posting on Facebook telling everyone we should set the local migrant hotel on fire. These are not political prisoners.

https://lordslibrary.parliament.uk/select-communications-off...

For the same reason you need to show ID to go into a bar/pub.

Did the Assange allegations fall apart? I thought he just hid until it became pretty irrelevant.

This is not at all what Palantir OR SamDesk does. SamDesk processes and sells data (eg, has agreements with various news/social media websites etc and converts that to a nice API of events). Palantir sells software to use all sorts of data, including stuff like SamDesk (hence why I suspect plenty of Palantir customers are ALSO SamDesk customers). They're complimentary rather than competing products.

"solved through technology" is an optimistic take on modern litter robots, in my experience ... I saw those as someone who has owned like 3 of them. Sometimes I realise I'm paying 500 bucks for an incredibly unreliable machine that gets slowly more smeared in poop over time instead of spending 30 seconds of my day to scoop some poop and feel quite silly.

Someone tested this, and it doesn't look to be saturated.

https://dylancastillo.co/posts/pelicanmaxxing.html

Simon made I think a very good argument for why it's still useful, if not the most robust benchmark in the world.

https://simonwillison.net/2026/Jul/16/kimi-k3/

At the risk of bringing up a controversial topic, in the midst of the Jason Arday furore last month when everyone was like "this wouldn't happen to a white professor", the person who kept coming to mind for me was Ariely. I became a data scientist in no small part due to his early work, and seeing academia totally incapable of punishing him is deeply disappointing.

Yup, you can still get one, as long as you're happy to pay 3k usd plus the 300 yearly cloud subscription...