HN user
Glad to see a benchmark outside of programming tasks. Presumably this one was not part of the training and shows the models performance (or lack thereof) on knowledge tasks.
Yeah on internal non coding benchmarks at my company, they are around last years models and don't hold up well.
Are "open" models exempt from that? Can imagine they used the same datasets.
this has been my tinfoil hat theory. Investors in US models might be supporting "open" models as a means to create FOMO for other investors to supply more cash to US model providers, which in turn increase their investment value. Package it so they can beat those "adversaries", equate that success with global power struggles etc etc. Seems to be quite effective.
Gonzo style / polemic would make a good filter on here.
As much as I dislike them [1], DSLs are an easy way to provide domain specific backpressure to your llm which might not easily be available in your language.
1- https://mikehadlow.blogspot.com/2012/05/configuration-comple...
Recently revamped my terminal setup after all IDEs have just gotten painfully slow to work with (the debugger + git integration in intellij was my last moat, but spend some time to learn nvim-dap + lazygit and it's excellent). AI has been immensely helpful here too to figure out the long tail of weird config gotchas.
Also thanks confirming the multiple cursor YAGNI for vim, could never wrap my head around needing it in the first place.
Eh it's not as black and white as you make it look. Colleague of mine is in ICE detention, because ICE acted on courts being held up in appeals so they can ignore his ongoing asylum case and deport him back to Russia. He followed the rules, had a work permit and everything, but did not matter in the end.
The green card interview snatching is also messed up, if your existing visa expired, while you were being processed, USCIS was understanding and it did not affect your application. (Processing time is slow, so that can happen). Now it's if your visa ever expires your Freiwild for ICE. They're technically not wrong with this, but they're essentially throwing the book at people getting a visum legally too.
Yeah, not to make light of the tests, but those degrees boil down to paying Goethe Institut to take their classes that prep you for the test.
Depends on your own taste for risk, should the knock off brand have worse QC: what big brand gets you is the ability to sue them should their products fail catastrophically or cause you harm:
https://www.geekwire.com/2019/lawsuit-ruling-dog-leash-purch...
https://www.bbc.com/news/articles/c1k2ydn1rz8o
It seems like the retailers can be held responsible should "ASDAS_A!kr" drop off the radar, but might still be easier to sue local.
(I know "local" companies still find ways to settle / weasel their way out of responsiblities, but at least you know where to reach them...)
Not in the field, but every time vitamin D studies come up I am reminded of the one that called out how current recommendations are based on faulty math (confusion on how to combine different sized studies confidence ranges ) and miss the mark significantly (and a lot of studies are based on those recommendations...)
Love plotnine when I switched over to python and great to see the project develop! But I have to admit I ended up switching to altair after all which has been my go to in python now.
It was a business trip, but yeah.
Parent was referring to the cost of hardware. I've had colleagues from brazil visit the US and go absolutely crazy at best buy to grab as much hardware as they could (laptops, nintendo switch, etc), because it's prohibitively expensive for them to buy that at home.
As an anecdote re Germans: A friend of mine did an Auslandssemester there and was surprised to see "No Germans" signs for some of the housing options. Always makes me chuckle as an example how "relative" xenophobia is.
Doing a western blot right takes a bit of practice and there are a couple failure modes you need to watch out for. Stuff like background "noise", smears, drifts can make it hard to get binary decision out of your experiment. E.g. antibodies are usually very very specific, but they can have impurities, unspecific bindings to other proteins etc which make interpretation harder. If they remove these from the advertised images you'll have a hard time comparing your own results to them. ESPECIALLY if they remove whole bands from the gel picture, which imho should be very much verboten.
Typically these catalogues have some numbers with regards to the antibodies binding affinity / impurities so you can have a general idea of what to expect, but having a clean image might mislead you into thinking that you did something wrong in your own setup. Seeing how wide spread it is, it's easy to imagine that their own lab is not run very "cleanly" and they have antibody contaminations in their gels, or issues with their own protocol that they're trying to edit out. Doubt that's the case, but it's really not a good look.
Looks like a manual override was triggered.
Very cool and love the animation, but does not seem to work well with beards :D
not the maintainer, but at first I thought it was python :D
yes, wondering if a simple pre/post llm processing would be enough?
urgh of course it has to be private equity. Really liked the product and did not mind paying for it...but not ready for the PE enshittification.
It is...? H-mart + Wegmans has tofu at ~$2.5 for a 400g block. The cheapest bulk pork is at $2.6, but most portions / cuts sell at $4.
not soooo much though. It's heavily subsidized for residential consumption, but industrial power rates are almost comparable to the US (depends on the state you go to etc).
it's funny how adding AI to notion actually made it a lot more usable. Most products force it on you, but here I feel like it's actually a massive benefit. It was hard finding content and using the filters felt clunky. (And the whole UI either in a browser or their app feels buggy + slow). But with their notion AI / MCP it's gotten super easy to get information in and out.
Soo can I put this on top of e.g. grafana?
Interesting that the MCP-Atlas score for 4.6 jumped to 75.8% compared to 59.5% https://www.anthropic.com/news/claude-opus-4-6
There's other small single digit differences, but I doubt that the benchmark is that unreliable...?
huh neat, somehow completely missed out on the rules/ + path filters as a way to extend CLAUDE.md
there's probably a more precise way, but if you're on uv:
rg litellm --iglob='*.lock'funnily enough the model switching is mostly thanks to litellm which dspy wraps around.
Hmm not quite what I meant. Sklearn has it's place in every ML toolbox, I'll use it to experiment and train my model. However for deploying it, I can e.g. just grab the weights of the model and run it with numpy in production without needing the heavy dependencies that sklearn adds.