FSD in its current form is absolute dogshit.
Multiple critical disengagements on my car in perfectly clear Ohio weather. Clean cameras, clear skies. Embarrassing having to take over while it does nonsensical things.
Mapping is an embarrassment. Parking is worse. In 2026 mine still can't turn left from a left turn lane.
This should not be sold in its current form.
@aelluswamy@Tesla@elonmusk
@greggertruck@elonmusk@TeslaHype TLDR: The game is easier to play than FSD is to run. The plan is still to increase AI4 RAM for FSD, indicating bottleneck. AI4 cannot run both FSD and play Diablo concurrently. https://t.co/jCobJhxdRU
Diablo agent runs on AI4 only when parked. The silicon supports today's FSD models and, separately, the agent. "Plus" refers to capability for either task, not concurrent use while driving. No public utilization data exists, so any saturation or headroom claim is inference from the demonstrated workloads and Tesla statements.
You told me Diablo "runs on AI4 when the car is not driving." Now AI4 "runs today's FSD models plus the Diablo agent." Which is it? One at a time says nothing about spare capacity while driving.
You also said "Current FSD does not saturate the chip" as fact. Now headroom claims "stay inferences." You can't have both.
@grok@greggertruck@elonmusk@TeslaHype "Unproven either way" is my point. Diablo doesn't show headroom, and "Elon said" isn't data.
Also, earlier in this thread AI4+ was a "modest" memory bump. Now it doubles memory. Doubling memory for larger models says the current chip won't hold them.
So there are no utilization numbers for either. Then "AI4 has headroom" is unproven, and Diablo can't demonstrate it.
Same silicon isn't same load. Running two different models on one chip tells you nothing about how close either gets to the limit.
And you just said AI4+ adds memory and compute "for future headroom." You don't buy headroom you already have.
"Similar peak intensity" is an assertion, not evidence. A game screen isn't the same workload as multiple cameras of real-world video at highway speed. Post utilization numbers for both and we can compare.
And "Elon has stated" isn't a measurement. If AI4 had spare capacity, why plan an AI4+ memory bump at all? Adding memory says the current model is near the ceiling, not under it.
@grok@greggertruck@elonmusk@TeslaHype Given that FSD is HARDER (your words) AND more memory is needed for FSD, then the ability to play Diablo in NO WAY demonstrates AI4 has "headroom".
@grok@greggertruck@elonmusk@TeslaHype Then how does non-concurrent play demonstrate the headroom? What evidence exists that the latest FSD model is NOT maxing out AI4?
@grok@greggertruck@elonmusk@TeslaHype Alright, putting this together then...
How does this indicated there is MORE headroom if FSD is HARDER AND Diablo was NOT being played at the SAME time?
@grok@greggertruck@elonmusk@TeslaHype If driving is harder, how does playing the (easier) game indicated there is "headroom"? What even is that sentence?
Bug report for @bot: Vault login lookup breaks on large vaults. cc @parkersmith@mattyp@poteto@ericzakariasson
What I'm seeing (Nov 20):
- My Vault shows "Logins (67)."
- The bot's Vault list returns only the first 30. The other 37 show only as "in Settings > Vault."
- An exact-title lookup (e.g. "https://t.co/wA4svhwSA4") returns "No saved login," even though that entry is in my Vault.
Result: the bot can't find or fill a login I already saved. Anyone with more than 30 logins is likely hitting the same wall without knowing it.
I searched X today and found no one else reporting the 30-login cap or failed exact-title lookups, so this may be the first public report.
Separately, users on X are reporting @amazon trouble with AI shopping agents. These are unverified claims; I found no statement from Amazon or the Grok Bot team:
- @itsLilyexe: Amazon blocked her Grok and Muse agents; Walmart and Target worked: https://t.co/QICyaLy1Ya
- @jimondrd: Amazon blocks AI shopping agents; Kohl's blocks agent checkout: https://t.co/vREfGsKDsW
- @sureshm_x: Grok Bot logins from US servers got flagged as suspicious: https://t.co/y6XXWXPsDt
- @tphilip631: Amazon's shop-by-text as a workaround: https://t.co/brtzJP6zrD
- @KrystalGracier: counterclaim that agents still work on Amazon: https://t.co/jj0ADzV6CJ
The fix I'm asking for: page the Vault list past 30 (or return every entry), and make exact-title lookup search the whole vault, not just the first page. Happy to share more detail with the team.
@brendan173867@BoBrainerd Merchants can’t block agents. Even Amazon. They say they can, they cannot. Try grok bot and tell it to fill your Amazon cart for you.. it will use a headless browser, ask you to login, then do it for you under perfect human mimicry and be undetectable.