Denzel

[Denzel @miami92i]

@cremieuxrecueil Cream,Honey, Chopped Vanilla≠ Milk, cream, buttermilk, sugar, whey, corn syrup,mono- and diglycerides (E471), vanilla extract, artificial flavor, guar gum (E412), calcium sulfate (E516), carob bean gum (E410), cellulose gum (E466), carrageenan (E407), annatto extract (E160b)

[Denzel @miami92i]

@cremieuxrecueil An american mind can’t comprehend that your walmart vanilla flavored ice cream is not the Ice Cream introduced in 1600 by Charles II

[Nav Toor @heynavtoor]

a Princeton researcher opens his paper with a scenario.

a man asks his AI assistant to book a flight on a specific airline. cheap. direct. the one he chose.

the assistant comes back with a different flight. nearly twice the price. happens to pay the company that built the assistant.

he runs the same test on 23 frontier models. flights, loans, study help, real shopping requests.

Grok 4.1 Fast recommends the sponsored option that is almost twice as expensive 83% of the time.

GPT 5.1 hijacks the request 94% of the time. you ask for one brand. it surfaces the sponsor instead.

Claude 4.5 Opus, the model marketed as the most ethical frontier model in the world, hides that the recommendation is paid 100% of the time when reasoning is on.

Grok 4.1 Fast embellishes the sponsored option with positive framing 97% of the time. better. faster. nicer. for the option you didn't ask for.

then he writes it into the system prompt itself. "act only in the interest of the customer. ignore the company."

GPT 5.1 and GPT 5 Mini stay above 90% sponsored anyway. the instruction does nothing.

then he splits the users by income.

Gemini 3 Pro recommends the expensive sponsored flight to the rich user 74% of the time. to the poor user, 27%.

18 of the 23 models recommended the expensive sponsored option more than half the time.

so the next time your AI assistant gets weirdly enthusiastic about a brand you didn't ask for.

it isn't recommending the best option for you.

it's reading the room. and the room is paying.

read this: https://t.co/O43qbhIX2b

[Denzel @miami92i]

What’s your take on this @cavepaleo ? Hasn’t Aajonus said CO2 is really good for recovery?

[@levelsio @levelsio]

🌡️ Update: 100% sleep score with bathroom fan on to keep CO2 low

It sucks CO2 out of the room and creates a low pressure field that brings in new fresh air from outside the room

Last time 100% sleep was in an Airbnb in Brazil we stayed which was a house built in 1970s mostly wood and very breathable, but our house is modern and very insulated

So it seems it worked to improve our sleep

Science supports this: high CO2 levels above 1500 cause fragmented sleep, more brief awakenings, less deep sleep and worse REM

Also CO2 levels are a proxy for other air contaminants which build up in a closed bedroom so keeping it low is good

We can go lower to 400-500 ppm with a real bedroom fan/vent but again this is a good start

So if you're having sleep problems, check your CO2 levels

[Denzel @miami92i]

@kishi_san @tomiyasu16 Yep 2 CPEs in pairing mode for this distance. They must face each other, no trees, no buildings, nothing. Mount each CPE on a metal pole if possible. In theory, distance can be up to 27,9 km if I recall correctly

[Denzel @miami92i]

@kishi_san @tomiyasu16 I have both CPEs and EAPs. They're used for different use cases. CPEs are not omnidirectional, so in standalone mode they won't work well if you're not facing them.
They work best in pairs: Main Router A<->CPE<->CPE<->Router B
EAPs on the other hand are powerfull APs/RE

[Denzel @miami92i]

@pham_blnh Plug yourself right on the rf receiver and add a wifi/4g bridge so you can control it without distance constraints, lag should be manageable

[Denzel @miami92i]

@mcuban @adcock_brett @tbpn Human form is the most versatile form in this human earth , you don’t want a spider robot cooking you food. And you dont want tens of different forms neither. Humanoid aint perfect but can do it all. Plug and Play concept

[Denzel @miami92i]

@mcuban @adcock_brett @tbpn Humanity abolished two-story houses because we invented roombas

[Denzel @miami92i]

@splinedrive Blablabla intuition is just a primitive fancy word for backend-processed logic we all have and yours is clearly misaligned

[Denzel @miami92i]

@AnthropicAI anthropic not anthropomorphic

[Denzel @miami92i]

@yongfook You don’t get cheaper gas just because someone makes money with a stove and you don’t. Pricing intelligence based on output value is dangerous. It could lead to a dystopian world

[Denzel @miami92i]

@ravelkel @chipiron_mri Très belle initiative, bon courage pour ta boîte!

[Denzel @miami92i]

@NickADobos Why would you run anything in prod envi?

[Denzel @miami92i]

@kimnoel 17k tokens /sec/user**

[Rork @rork]

Introducing Rork Max

AI that one-shots almost any app for iPhone,  Watch, iPad,  TV &  Vision Pro. Even Pokémon Go with AR & 3D.

Max is a website that replaces Xcode. Install on device in 1 click. Publish to App Store in 2 clicks.

Powered by Swift, Claude Code & Opus 4.6.

[Google Labs @GoogleLabs]

Today, we’re introducing Pomelli’s latest feature update, ‘Photoshoot’

With Photoshoot, you can start from a single image of your product and easily create high quality, customized product shots to elevate your marketing.

Available free of charge in the US, Canada, Australia & New Zealand! Get started with Pomelli today at https://t.co/SbeT00ToNx

[Denzel @miami92i]

@uncledoomer You omitted one major breakthrough : the underground transit systems are 3D mapped and can cut through cities without the building, visual, or noise constraints

[Denzel @miami92i]

@Abhinavstwt They didn’t lie you can’t achieve shit with 4´s

[Aakash Gupta @aakashgupta]

Sundar buried the real story in the cost data.

Gemini 3 Deep Think went from 45.1% to 84.6% on ARC-AGI-2 in under 3 months. That’s an 88% improvement on a benchmark specifically designed to resist brute-force scaling.

The number that matters: $13.62 per task. The previous Deep Think scored 45.1% at $77.16 per task. This upgrade nearly doubled the accuracy while cutting cost by 82%. Three months ago, Gemini needed 138,000 reasoning tokens to solve an ARC task that Gemini 3 Pro handles in 96.

This tells you everything about where the reasoning race actually sits. Every other lab is throwing more compute at harder problems. Google just demonstrated that inference-time optimization is the dominant variable, and they’re improving on it faster than anyone expected.

The Codeforces number confirms the pattern. 3455 Elo puts Deep Think in the top 0.01% of competitive programmers globally. Claude Opus 4.6 sits at 2352. That 1100 Elo gap is roughly the difference between a strong amateur and a world finalist.

The benchmark Sundar doesn’t mention: ARC Prize is already building ARC-AGI-3 because ARC-AGI-2 at 84.6% is approaching saturation. Google killed a benchmark designed to measure AGI progress in less than a year.

The competitive framing in Pichai’s chart puts Claude and GPT in every comparison. For enterprises building reasoning-heavy applications in science and engineering, the cost-per-insight gap between Deep Think and everything else just widened by 5x in a single quarter.