skradacz

@mikepat711 i've got claude subscription but only use it for claude code, while for research and web search i use grok. what do you use work helper/cowork for?

This is one of the two biggest reasons why I still end up not using grok that often. The second is that it lacks features like Claude’s work helper/cowork, or perplexity’s spaces/perplexity computer. Seems like the real emphasis is on imagine, which is cool but mostly useless for someone like me. Hoping to see them add stuff like this soon. Today I’d say I use Claude 80% of the time, Gemini 10% (because it’s integrated with chrome), and Grok the other 10%. Just started using comet though which is probably replacing Gemini. And I am a person who actually wants to use Grok over all others. But the features others have are just too useful for now.

@mikepat711 i trust your opinion more than the academy's lol

@mikepat711 thanks, i'll watch it

This is fair. Paul Thomas Anderson’s best movie in years. And one of his best ever. Also one of Sean Penn’s best roles. If you started it and turned it off because it seemed like some woke trash, give it another shot. The movie isn’t what it seems in the first 20-30 mins. It’s a banger.

@GibsonHimself @mathemagic1an @grok here you go grok.com/share/bGVnYWN5…

@mathemagic1an I don't know what this mean and can't afford to have @grok explain it

Transformers are Turing complete and can be trained to run arbitrary programs Turns out you can embed a relatively efficient assembly interpreter in the forward pass. This allows the LLM to execute deterministic code at inference time in its own weights, no sandbox

i honestly don't think he is contradicting himself. i think he's stuck in a worst possible scenario: he needs some kind of driver assist and his unit volume might be too small for Tesla to bother letting him license FSD. "LiDAR has no on cost" - here he was comparing how small cost is vs trying to do what Tesla does training wise "but actually it does" - it sure does "but well it doesn't because we save the cost thanks to our chip" - now he's comparing not vs Tesla, but vs Rivian's older models. i don't know if the math works out, but it's possible that what he says is true. "And you need more sensors" - again, you do if you don't have the data and compute "but actually you may not in the future we will see" - in 10 years they might get enough data and compute might be cheap enough that it will be easy to catch up to where Tesla is now i'm more confused by what is saying regarding edge cases

@mskradacz As I said, I don't understand because he is contradicting himself. LiDAR has no on cost but actually it does but well it doesn't because we save the cost thanks to our chip. And you need more sensors but actually you may not in the future we will see...🤪

I have been doing this for my whole professional life, and I do not understand what he says. It's incoherent. "supplement that with LiDAR [...] allows us to train the vision models faster. And the incremental cost to do that is relatively small." LiDAR is extremely expensive. One LiDAR is roughly the cost of the compute itself. It almost doubles the hardware of the system for every car. Add the RADARs and you are clearly two times more expensive than the Tesla system. Having LiDAR on every car is not necessary to train your vision models faster. You need some LiDAR data to kick start your auto-labeling, but when you have this you can create labels from the video streams alone by using frames "from the future" (use frames n+1, n+2... to improve the labels of frame n, n-1 etc...). "any of the incremental costs that would have been there on its own is offset by the fact that we brought inference in house" So yes, there is an incremental cost, but actually there isn't because we make our own chips and don't buy from NVidia anymore. So which one is it?🤔 "in like the infinite long-term [...] once the models are very very robust you could have less cameras or you may be able to get away with less radar." Well, here he recognizes that when the models are good enough, you can get away with less sensors. He does not quite say vision only, that would be a bad look. This was my point all along: Rivian relies on sensor fusion to patch the gaps of their under performing vision models.

@nafofafo46877 @christianpet_ @elonmusk

@christianpet_ @elonmusk Fuck Tesla and fuck you you racist cunt

For years leftists have been harassing Tesla locations around Sweden. Now we caught one on hidden camera. An IF Metall union activist calls violent attacks on Tesla and @elonmusk “GREAT” — says activists have been “too nice”. When confronted? She RUNS. Mob tactics because @Tesla won’t sign a union deal — despite offering BETTER conditions than union norms.

@BigSquid0508 @DenisGajcew @Pirat_Nation congrats, 7 retards liked your post so far because they have no idea you made it up

@DenisGajcew @Pirat_Nation They do actually if we go by both the books and games, it's just very rare

CD Projekt Red co-founders have been named in a reopened Polish investigation into the 2018 death of a 54-year-old Polish woman on a farm in Janov, Czech Republic. The woman died during a shamanic ceremony involving ayahuasca, a psychoactive brew that is illegal in many countries. Polish prosecutors reopened the case based on new evidence, including materials mentioning the founders due to property connections: Kiciński co-owned the Janov farm where the ceremonies took place and reportedly participated in some ayahuasca rituals earlier, while Iwiński owned a nearby property used for similar events. No charges have been filed against them. The investigation remains ongoing as of March 2026.

@BigSquid0508 @DenisGajcew @Pirat_Nation none of the books mention it. go on, find a single quote

@DenisGajcew @Pirat_Nation Habe you ever read the books? The school of the cat literally gets mentioned in both the lady of the lake and the tower of the swallow, a school that PUT GIRLS THROUGH THE TRIAL

@ardacebie @MdeZegher @TeslaCharging obviously external units, that's why "sit mostly idle outside of holidays" is mentioned

@MdeZegher @TeslaCharging Could you elaborate on what the “units” are, Max? Are you referring to the AI4 inference on the vehicles which will be active during Supercharging sessions, or will these be external units positioned on Supercharger sites?

@electricfelix @LiveAir4real ?

@LiveAir4real no way RDW signing off on this next week

Just had another FSD Supervised ride in Hamburg🇩🇪 is Tesla really about to kick off something HUGE in Europe?🤔

@Vrele_Gume @Vrele no they didn't

Tesla pokazala humanoidnog robota nove generacije: Optimus V3 spreman za masovnu proizvodnju vrelegume.rs/tesla-pokazala… via @Vrele gume

@mikegapinski Nie rozumiem co widzę. To jest Autosteer czy prawdziwy FSD? Jak go włączyłeś w PL?

She rips, we’ve got it ladies and gentlemen. Mad max profile is really MAD, it goes between cars like a BMW driver on a 3 lane highway well above the speed limit. That’s a bit much for me, I’ll dial it down. Granular speed control still available and works! All the EU localisation are there in 2026.2.9.1 (hands off mode, a message that notifies before lane change happens) Happy days!!!

@EstherRebers @Tesla_Calle @Tesla @EvaFox @SERobinsonJr @Kristennetten @mayemusk @wholemars @SawyerMerritt @EvasTeslaSPlaid "a very different experience from the drive at Tesla Düsseldorf" how so?

FSD Supervised in Europe: @Tesla_Calle and I had a FSD Supervised ride along drive from @Tesla Tilburg (the location we all know so well in Europe) in The Netherlands just now! Such a great ride with many interesting moments that I was able to capture on film. Also a very different experience from the drive at Tesla Düsseldorf. We are so impressed! More great videos soon! @teslaeurope

@TruthPointUSA @wholemars Hahahaha the driver assistance comparison is so stupid

@wholemars Na. Hybrid is better. Even at 9 per gallon.

Tesla dot com

@RuffBuffet @wholemars yeah, but it's not the same tokens? your local model running on mac studio is much less intelligent than grok-4.20

@mskradacz @wholemars I had both write the same article and then timed how long it took to output it.

It’s not really a binary outcome. Many simple use cases like summarization, rewriting, etc will be served by small local models, especially as local inference power increases But the long running agents developing software aren’t going to be running on an iPhone.

@RuffBuffet @wholemars Weird comparison. Not all tokens are equal

Future iPhones will have some impressive inference compute capabilities (a Mac Studio can spit out 3,000 tokens faster than Grok can today), but we're still going to need datacenters for training and some inference loads. Not everyone is going to want a 3% battery hit on their phone just to answer a question.

@JTLonsdale just say you're deeply closeted

In CA this month, we’re literally required as a venture firm to ask each of our CEO’s in a survey if they are gay. Meanwhile today in Texas the voting machine is like, Y or N, “Ban gender nonsense in K-12?” “Ban sharia law?” 🇺🇸

@bexxxyss @westerosies He makes no mistakes

@westerosies where is baelorrr

First look at the Blooper Reel for HBO’s ‘A KNIGHT OF THE SEVEN KINGDOMS’ ⚔️ The reel will be included in the 4K UHD Steelbook releasing on June 15.

@ChrisLancasterX @PTrubey @YAppelbaum how would spacex do this without ukrainian whitelist?

@PTrubey @YAppelbaum That's BS! Musk could have turned off Starlink for Russia at anytime in the past two years. X was flooded with reports of Russians using Starlink to guide their drones into Ukrainian civilian vehicles and buildings.

"The SpaceX team behind it received clear instructions from its bosses: “‘No limits. Take off the gloves; use Starlink for anything to help Ukraine.’”"