nikcub

* real programmers write assembly, not FORTRAN

* real programmers manage memory, it's a craft

* real programmers don't drag and drop

* real programmers don't use intellisense

* real programmers don't need stack overflow

* real programmers don't tab-complete

* real programmers don't need copilot

* real programmers don't use llms <- you are here

> If a tree falls in the forest...

I hope this is a pun on the content management system used to publish OP. It's forester[0], written in OCaml and parses TeX-like .tree files into semantic XML which uses browser XSLT to render the HTML.

View source on the page to get an idea.

Reminder of what the idealised web promise from decades ago was. Long gone. Very apt.

[0] https://www.forester-notes.org/index/index.xml

the most cited is terminal bench 2.0, but its also plagued by cheating accusations and benchmaxxing.

somewhat remarkably, claude code ranks last for Opus 4.6 - which may say something about cc, or say something about the benchmark

[0] https://www.tbench.ai/leaderboard/terminal-bench/2.0

Stunning results at the top of the field. Some interesting takeaways on both fuelling and shoes.

Maurten spent months working with Sawe and other runners getting their gut capacity trained so they could absorb and burn 100 carbs per hour[0][1]

> The Maurten research team was embedded with Sawe’s team in Kenya for 32 days across six trips between last and this April. They were training his gut to absorb that load by mimicking race-day protocol in training. The hydrogel technology they have developed over the past 10 years now allows athletes to absorb 90–120 grams of carbs per hour without GI distress.

Second is the shoes. Adidas Adizero weigh 96 grams[2] with new foam tech and new carbon plates

Nike and INEOS spent millions over years to get Kipchoge to a sub-2 in artificial conditions, and now the elite end of the field are knocking that barrier out in race conditions. Unreal.

Running tech and training have been revolutionized in the past few years.

[0] https://marathonhandbook.com/sebastian-sawe-arrives-in-londo...

[1] https://www.instagram.com/p/DXmvAUvkWaq/

[2] https://www.runnersworld.com/uk/gear/shoes/a71129333/sabasti...

edit: correct :s/calories/carbs thanks

Airlines are down there amongst cinema chains and video game retail stores in terms of being terrible businesses

Amazon and Google get discounts because they bring more than just cash and help solve a very immediate problem for Anthropic

Great position to be in if you're Amazon and Google

Google cloud also need to be able to offer Anthropic models on Vertex otherwise they just won't be competitive.

Microsoft is in the same boat with Azure.

compartmentalize. I do development and anything finance / crypto related / sensitive on separate machines.

If you're brave you can run whonix.

The issue is developers who have publish access to popular packages - they really should be publishing and signing on a separate machine / environment.

Same with not doing any personal work on corporate machines (and having strict corp policy - vercel were weak here).

and then released a model that didn't really leave a mark with code performance

it's not dollars it's X bucks

> hand out free top end model coding access for a bit and you'll very rapidly have developer data

They tried this - grok was free on openrouter for a while

knee-jerk is that it's weird, but makes sense:

* X will have a total of ~2GW of GPU sometime this year largely not doing much outside of 'grok is this true'

* despite no longer being in vogue with consumer devs Cursor still has a lot of developer data that can assist in building a model

* Cursor have decent enterprise relationships (while for xAI it is ~zero) and that's where the real revenue for llms + agents is

* Cursor are paying retail for tokens and competing against the frontier model co's who are also their suppliers. Not sustainable (hence their in-house composer model).

* Cursor the product covers the gamut from lovable-style prompt-to-app, an IDE, cli and bugbot

* X are using "x bucks" to pay for a potential later acquisition which are arguably overvalued based on the space x IPO hype

Option there to give X a window to make it work, otherwise walk away with a $10B breakup fee for access to it's data

Claude Code defaulting to a certain set of recommended providers[0] and frameworks is making the web more homogenous and that lack of diversity is increasing the blast radius of incidents

[0] https://amplifying.ai/research/claude-code-picks/report

he's completely ripped that post from the person who originally found it on breach forums. absolutely shameless.

https://x.com/DiffeKey/status/2045813085408051670

Department of War timing on picking fights couldn't be worse

ugh, strong ai slop vibes with comments like this

a) llms are good at writing typescript

b) typescript fixed a lot about javascript and is somewhat decent

c) multiple fast and performant runtime engines

d) deployment story is php levels of easy

that's it.

interesting that you only need ~150 stars on a project for it to be in the top 1%

the fact that somebody was able to fork it and remove behaviour they didn't want suggests that it is very open

that #12446 PR hasn't even been resolved to won't merge and last change was a week ago (in a repo with 1.8k+ open PRs)

This is analogous to when Google launched Gmail with 1GB of storage and then a bunch of third-party apps cropped up that took advantage of it to use it as a generic online file storage drive.

There was GMailFS[0] and Gmail Drive[1] - this is before S3, dropbox, and a time where web hosting would give you ~10MB or so of space.

Google updated their ToS and shut down accounts using their service in ways they weren't intended via these apps - because obviously the 1GB of storage was a loss-leader into Google's ecosystem (and it worked)

Same thing today - "unauthorized" third parties taking advantage of a loss-leading[2] deal - complete with similar trademark violations to boot[3].

Google have more cash to burn in the AI race so can be more forgiving today in how their codex plans are used. Anthropic are still a private company and can't.

[0] https://handwiki.org/wiki/GmailFS

[1] https://techcrunch.com/2005/07/31/profile-gmail-drive/

[2] it's a big q just how large a loss leader the max plans are considering a fixed harness, prompt caching etc. but point still stands. you're getting up to $5k of RRP tokens for $200

[3] Clawd Bot -> OpenClaw

the $20 pro plan would also have double offpeak limits - just set it to sonnet and you'll get a reasonable level of output

thank you!

I can't find a github or email for Hannah - if you're reading this i'd like to add Australian energy price data via Open Electricity[0] to the data (reach out via my profile)

[0] https://explore.openelectricity.org.au/

chatgpt use should be in the default set since energy use of ai is so often in the news now - and more often in social media

so we're all going to hold onto sequoia like we did snow leopard. only reason i'm not buying a new mac at the moment is because it would force me to upgrade.

the situation is absurd ..

fwiw switching to the sequoia beta channel in system settings killed the nag notifications for me (I believe the profile as defined in OP will stop all updates - which you probably don't want)

> Claude Code this morning was about to create an account with NeonDB

I had the same thing happen. Use planetscale everywhere across projects and it recommended neon. It's definitely a bug.

from my understanding Anthropic are now hiring a lot of experts in different who are writing content used to post-train models to make these decisions and they're constantly adjusted by the anthropic team themselves

this is why the stacks in the report and what cc suggests closely match latest developer "consensus"

your suggestion would degrade user experience and be noticed very quickly

prompt caching - big part of the reason why they can economically offer claude code plans. one of the ant team explain it here:

https://x.com/trq212/status/2024574133011673516

> they say they use heuristics and usage patterns.

cache hit rate alone would stand out

it's gone a bit unnoticed that they've stopped support for response prefilling in the 4.6 models :/