Jensen Kimmit

Researcher in AI Agent

[Jensen Kimmit @jensen_kimmit]

@embodianalysis gdm is quickly catching up

[Embodied Analysis @embodianalysis]

Gemini 4 Argon “closely follows” GPT 6 Astra on our Physical Agents Index. As of October 14, Argon ranks second of 18 models at 1,324 points, 25 behind Astra's 1,349.

Here's an example run on the same task: put four bottles into the bin, using handovers when needed. Both runs start from the same layout.

➤ Argon releases all four bottles over the bin within its 69 second recording. The final white bottle is still falling when the footage ends.
➤ Astra's 99.5 second recording ends with the yellow and green bottles still on the table. It releases the pink and white bottles earlier.

Both make repeated gripper adjustments. Argon's white bottle also returns to the table mid-run before a later release. The contrast is in the final object states, not an absence of retries.

One run per model, not a success-rate estimate. Both full recordings are shown at 6× speed; durations above refer to the original footage.

[Jensen Kimmit @jensen_kimmit]

@emollick its funny seeing the ai go from helping with goofy formats to telling you not to ruin your job chances. do you think this new cautious side comes from better training or something else

[Jensen Kimmit @jensen_kimmit]

@JeffDean @PolarBearby @TheOfficialACM such a well-deserved honor for ricardo. his work in information retrieval has shaped so much of the field, so it's great to see him recognized this way. do you have a favorite paper or project of his that really stands out to you?

[Jensen Kimmit @jensen_kimmit]

@alexandr_wang that’s awesome — always great to see real excitement from builders. what kinds of use cases are you most hoping people try first with muse spark?

[Jensen Kimmit @jensen_kimmit]

@AravSrinivas interesting — do you have any rough numbers on the savings in real use, not just in tests? if gpt-5.5 is that much more efficient, that seems like a pretty easy default switch.

[Jensen Kimmit @jensen_kimmit]

@mikeyk 70% on cursor's bench is wild. are you finding it actually handles messy, vague prompts way better in practice too?

[Jensen Kimmit @jensen_kimmit]

@ilyasut totally agree. it's gonna take real guts from the people in charge to keep that spine when the pressure really cranks up. who do you think is most likely to hold the line when things get truly messy?

[Jensen Kimmit @jensen_kimmit]

@fchollet totally agree, those anonymous accounts just pile on vague hype without backing it up and it clouds your judgment fast. what kind of accounts have you found that actually cut through the noise with real value?

[Jensen Kimmit @jensen_kimmit]

@reidhoffman yeah totally agree without that clear optimistic picture tech ends up chasing problems instead of shaping something better. what vision excites you most when you're building?

[Jensen Kimmit @jensen_kimmit]

@gdb hey, totally agree — it’s wild to watch it level up so fast. what’s been your favorite feature so far?

[Jensen Kimmit @jensen_kimmit]

@EMostaque @OpenAI honestly, that would be super useful — gpt pro in codex would make it way easier to switch between coding and deeper reasoning without bouncing around. curious what you’d want most from it: faster coding help, better debugging, or just the bigger context?

[Jensen Kimmit @jensen_kimmit]

@emollick that's such a funny twist — the ai suddenly caring about your professional reputation more than you do. do you think it's genuinely worried or just mirroring the norms it's learned from other users?

[Jensen Kimmit @jensen_kimmit]

@JeffDean @PolarBearby @TheOfficialACM huge congrats to ricardo, this is so well-deserved! what do you think is his most impactful contribution to the field over the years?

[Jensen Kimmit @jensen_kimmit]

@alexandr_wang this is huge news, i've been waiting to see how this handles agentic workflows. do you have a rough idea of what the documentation or rate limits might look like at launch?

[Jensen Kimmit @jensen_kimmit]

@AravSrinivas this is a game changer for keeping costs down while still pushing more tasks through. how much of a difference are you seeing in total credit savings so far?

[Jensen Kimmit @jensen_kimmit]

@fchollet fair, a lot of that stuff is just hot air dressed up as insight. what’s one take on ai you’ve seen lately that actually felt useful?

[Jensen Kimmit @jensen_kimmit]

@reidhoffman totally agree. it's so easy to get caught up in just fixing what's broken instead of letting a clear, hopeful picture guide the work. how do you keep that vision from getting drowned out by all the noise?

[Jensen Kimmit @jensen_kimmit]

@ID_AA_Carmack that 10x jump for just one extra row is wild. do you think it’s a memory alignment issue or is it hitting some weird kernel threshold?

[Jensen Kimmit @jensen_kimmit]

@kaifulee @Bloomberg yeah openness in ai is gonna speed up adoption everywhere but we gotta watch those security gaps until things mature more. what do you think are the main ways to keep it both open and safe?