Pythonics

@tunguz @geteviapp Has to be rage bait. This was tool gen output, if it opens up the Python interpreter to produce the comment it’s not the image gen model brother 😂

@geteviapp Yes

I tried making an infographic using the GPT-image-2. Lots and lots of visually unacceptable artifacts. :/

Well, that was a very casual release Opus 4.7

Honestly, H&R Block rules. TurboTax is the worst.

@mattpocockuk 5.4 xhigh in codex always asks me 100+ questions. It’s very tiring.

/grill-me just asked this guy 139 questions New record

So many people seem to be trying to use AI to work a lot harder, and bragging about it. Why are you bragging about your complicated setup that requires you to work harder? Have you tried just having the models do the work well? It’s out there man. Get out of the loop!!!

Alexa is really terrible overall, but the final straw might be that occasionally, when I thank it, it lowers the volume on the device and *whispers* a contextual “you’re welcome” back to me.

Just totally radioactive. These posts are not well advised 😬

gpt-5.4 xhigh in codex is the first LLM that can solve my embedded needs. Over a few Ralph loops, it delivered an efficient firmware for low latency USB cam streaming over its own soft AP, cutting our BOM cost down ~40% (SoC & power impact). ~$4k spend w/ dev shop failed, btw

@gm_mertd @Cloudflare @modal @northflank @flydotio @vercel The sprites can’t do long running agents in my experience, fails around the 3 hour mark intermittently (sprite just disappears)

I propose a new benchmark for sandbox providers: If CC or Codex was able to run it and integrate it by itself. Here are my results so far: 1. @Cloudflare : no 2. @modal : yes 3. @northflank : no 4. @flydotio Sprites: yes 5. @vercel : no 6. hopx: no

If you’re doing embedded work and researching components, design, firmware, etc. always always include a research path that has the LLMs search in Chinese. Night and day difference.

@dexhorthy Always in the dumb zone by default?? That’s a tough foot gun. Was he using the coding agent he’s built? Weird to me that the “smart zone” isn’t widely known about at this point. Poor guy

today I watched a coding agent founder try humanlayer for the first time and he dead ass had 260k tokens of mcps turned on

@BryanOnel86 They’re hiding the new auditors they’re using as they promise customers that the auditors will be different than those listed in the “reports” in an attempt to prevent churn. That, plus free stuff. The brand is so totally radioactive they don’t have a hope of shifting blame tho

Friendly announcement that Delve still lists Glocert, Sensiba and Aprio as their auditors. rsa.delve.co

@shl It’s going to be really awkward when I ask it to summarize the performance review it issues me

AI is going to go from your assistant to your boss

@BrendanFalk @NotionHQ Lmao but have you used notion though?

I feel like @NotionHQ is better suited to this new era they are describing. Linear's value prop is the speed of a human creating/managing tasks. If agents are now doing the work, I'd prefer to have them managed where 90%+ of my company knowledge already lives.

Simple sauce on effective automated background reviews. As this varies so much from codebase to codebase, engineering orgs on the frontier may have seniors deployed to “tune” the harness for some period of time, revisiting on some interval. Harness tuning is valuable.

@_colemurray Anything special you did for the fine tuning process? Mixed results for us leveraging previous review comments & diffs. I mean, better than coderabbit, but that’s a terribly low bar.

my AI code reviewer, i hand-tuned, flagging all the same things i would've in a PR 🥲

Opus 4.6 for those first 100k tokens on a really small task is still king man

@thorstenball Stick to raising agents, big dog

Something something models something something take the xhigh road. Is this anything?

Going to set up a recurring Claude job that reads my apple health data & notifies me if I may be getting sick. This one surprised me (was feeling pretty fine this morning) and is the absolute worst timing. I feel like an asshole for potentially spreading it as well. I don’t think my Apple Watch does this on its own , but I wish it did. @bryan_johnson are these actual indicators or is this a fast path to a type of psychosis?

Wait really high fevers can cause brain damage? Is that true? I really can’t take anymore of that

103.7 fever came on really fast, was feeling basically 100% a few hours ago. Big launch coming up… what should I do man?