@aarondfrancis @BenMcKayDev Looked it up and I'm wrong!
reduced fat milk is only like 1.5-2x whole, but 2% is actually much less than whole.
whole milk sales are on the rise
English
Sam Snelling
2.7K posts
Sam Snelling
@snellingio
building https://t.co/5Sf1CVIUTj
Oklahoma City Katılım Temmuz 2009
723 Takip Edilen1.3K Takipçiler
@aarondfrancis @BenMcKayDev whole milk being the regular milk actually is a bit surprising to me, I would've guessed 2% was the regular?
I think 2% sells at like 2-3x all other cow milks
English
When the barista asks what kind of milk I want in my latte and I say "just regular" I feel like God's bravest soldier in the fight against alternative milks
English
@aarondfrancis @elithrar are we hitting any safety blocks using the words "reverse engineering", are we phrasing in a certain way, or are the models finally smart enough?
English
I am similarly deep down the rabbit hole of reverse engineering undocumented embedded protocols.
I have a binary diff for a 50+ year old Bosch engine mgmt protocol + 64K EPROMs to write to.
Learned a ton throughout this via GPT 5.5.
Aaron Francis@aarondfrancis
Codex is currently trying to reverse engineer the Amaran Bluetooth mesh network protocol. It told me to buy this board, now it's flashing it with something I have no idea what it's doing but it's doing a great job
English
capability doesn't matter if you don't have access to the model
Lisan al Gaib@scaling01
> "I can drive 4x cheaper to the next city with my lamborghini compared to your helicopter" > "great, now get me to that island with your lamborghini" price/performance doesn't matter if you are capability locked
English
Sam Snelling retweetledi
🚨 Ongoing supply chain attack on Composer packages! We just found multiple laravel-lang/* packages compromised on Packagist (lang, http-statuses, attributes). Payload runs at autoload time. At least 50 package versions were compromised.
If you installed a compromised version, the malware already executed. Pin to a clean COMMIT (not version) and rotate secrets immediately.
If your lockfile already had an older commit from before today, you are safe. But you should not update at the moment.
English
Sam Snelling retweetledi
one thing I'm noticing is that all of these RL'd models inputs are scaling up like crazy.
I think there is a ton of unnecessary file reading & searching happening inside the context window.
my guess is it's partially harness related, and partially post training.
Adam Holter@AdamHoltererer
English
@ShiroLinHime yes, I said something super similar about the desktop app transition
x.com/snellingio/sta…
Sam Snelling@snellingio
from the product / engineer side, I totally understand how this decision got made, and even agree to some extent from the user side, the experience sucks and it continues to erode trust around google
English
@snellingio Streamlining gemini-cli's complex configs makes sense, but agy 1.0.0 feels half-baked. They should've polished it more before launch. I trust it'll improve with iterations, but this messy transition is destroying user goodwill.
English
My honest feedback on antigravity-cli:
0. Requires re-auth every launch despite being logged into the IDE.
1. Unreadable UI rendering during execution.
2. Zero UI hints for costs, skills, MCP, context, or current paths.
3. No auto-model allocation; manual switching is clunky.
↓
English
@ShiroLinHime yeah it seems to be missing most of the gemini-cli features as far as I can tell? idk why they went this direction it makes no sense to me.
English
@snellingio Thanks for the feedback. I just prefer how gemini-cli let you toggle YOLO mode with Ctrl+Y at any stage of task execution.
English
Sam Snelling retweetledi
I have done this 100 times I swear.
Robert Bye@RobertJBye
Me: why isn’t my MacBook charging Also me:
English
@ShiroLinHime 6. `--help` shows `--dangerously-skip-permissions`
I hit rate limits before hitting usability issues personally
English
4. Random freezes with no way to exit.
5. Non-CLI mindset for plans/artifacts—makes planning & execution tedious.
6. Couldn't find a YOLO mode.
7. I'd list more flaws, but your stingy quotas ran out before I could even use it. 😮💨
I missed gemini-cli. It was way better.
English
@adamwathan @vazuzu_varun i totally get how it’s potentially problematic, but my kids and i had plenty of fun having ai dress us up in silly costumes like chickens and dinosaurs.
i don’t know / have strong opinions on where lines need to be drawn, but the current refusal process feels wrong
English
Shout out to Grok for being the only model that will let me create images of my kids with face tattoos 👊🏻
English
I got it on the discount, and my review is:
- it's passable
- in the same bracket as minimax, glm, kimi
- build cli is both better in some ways, and worse in some ways
- even at $99/mo, there are much better deals out there
- you're not missing out imo
Wes Bos@wesbos
Grok's claude competitor is coming Grok build is only available to $300/mo SuperHeavy users at the moment. Anyone tried it? I've been very impressed at the speed/cost/quality of the latest xai models - specifically the speech and image ones.
English
Cursor is doing a lot of things right it seems
Artificial Analysis@ArtificialAnlys
Cursor with Composer 2.5 is the cheapest agent scoring above 60 on the Coding Agent Index at $0.07 (standard) and $0.44 (Fast) per task. Higher-effort variants — Claude Opus 4.7 (max) in Claude Code (66, $4.10) and GPT-5.5 (xhigh) in Codex (65, $4.82) score above at ~10x (Fast) to ~60x (standard) the per-task cost
English
Strong use case for Gemini here! looks great
Logan Kilpatrick@OfficialLoganK
Gemini 3.5 Flash ranks #1 on Automation Bench (from Zapier), beating every other frontier model at a much lower cost
English
I really dislike reasoning models. I hope this is just a stepping stone to better data.
idk exactly what my definition of model intelligence is, but it is probably something normalized (time, energy, latency, price, <something>).
Lisan al Gaib@scaling01
My definition of model intelligence has been very clear over the past two years. For me the sign of an intelligent model was always good results with as few resources as possible, which is why I was a big fan of Sonnet 3.5/3.6 and Opus models. These models would just get things and one-shot problems "without thinking". On the other hand I really disliked reasoning models from o1-preview up until o3, because it just wasn't worth it back then and felt like inelegant brute-force slop. You would get slightly better results for 10x the cost. Later from GPT-5 up to GPT-5.2 the reasoning budgets exploded from a few thousand tokens to 50-100k tokens. Since then reasoning efficiency has only improved, and we are now living in a world where GPT-5.5 and Mythos get insane results with very low reasoning budgets and where higher token budgets feel worth it. I think part of this is also that models nowadays know how much reasoning to spend on each problem. So when you set reasoning effort to xhigh it doesn't think for 100k tokens on a very easy problem just for the sake of the xhigh setting. (but personally I still use medium thinking budget like 90% of the time and will only go up to xhigh when the tasks have a high enough skill ceiling. it's overkill to use xhigh for everything)
English
yeah years of bad habits and hygiene are finally catching up to ecosystems.
David Cramer@zeeg
"update deps because theres a new version" is the worst thing the javascript community ever did to the world theres literally no semver compatible at all in the ecosystem, its full of spaghetti slop code (pre AI slop!) and deps cascade-change every day of the week its an absolute mess, and its no surprise that its constantly being exploited
English
Drake with the perfect 4o impression here. 10/10
Kurrco@Kurrco
"Normalize glazing the bros" — Drake Drake and Kevin Durant link up in a new ad for KD's 19th signature Nike sneaker 👀
English