Matthew Shen

compsci and bio @mcgillu

[Matthew Shen @flyme2bluemoon]

When I first heard that OpenAI was going to work on their own version of an always-on personal agent, I was quite excited. grok @bot had just dropped at that point, and I had quite a pleasant experience playing around with it.

now that openai dots is here, I'm much less excited than I thought I'd be. of course, the elephant in the room is that it's only available in the Pro plans and not in the Plus plan for us mere casual users. meanwhile, grok bot is available starting from the $20 cursor pro plan, while @Muse is even free! it would have been really nice to see dots included in the Plus plan, even if that meant a weaker model like 6.1-sol instead of astra and/or usage limits.

it's also a shame that they didn't implement anything like Muse's security architecture. Meta built Muse around the Muse Secure VM, and it would be nice if this became a common design principle across all personal ai agents.

however, despite all of this, I still really want to try out dots (Pro 100 plan, i see u 👀). Astra's computer use capabilities are second to none, and I think that's really going to set dots apart from the rest. I also think that dots and grok bot's implementation, where you have a team of agents, each with their own specialty, makes a lot of sense for this product category.

I just hope that the OpenAI team continues to cook and finds a way to make this product more secure and accessible (my wallet is shaking in fear at the prospect of a CA$136/month subscription).

https://t.co/HtMqgxo5J8

[OpenAI @OpenAI]

Introducing dots, powered by GPT-6 Astra.

Remarkably capable, always-on agents built to handle everything.

[Matthew Shen @flyme2bluemoon]

@poteto @bot is it possible to make the bot wake up when it receives an email without using a routine?
also, having +tags like gmail to route to different bots would be super dope!

[Token Gremlin @TokenGremlin]

I'm a traitor who switches between being an OpenAI fanboy and an Anthropic fanboy every single week.

[Matthew Shen @flyme2bluemoon]

@elonmusk @elshayib_ yeah, it definitely seems like grok bot usage has gotten worse in the past couple of days, perhaps due to opus 5.5 being more expensive, but it really sucks 😔

[Matthew Shen @flyme2bluemoon]

@ns123abc compute is solved, getting access to colossus is not

[Matthew Shen @flyme2bluemoon]

@elonmusk @bot @SpaceX spacexai really cooking across the entire product lineup! would love to finally see the subscriptions unified tho (I know it's coming but hopefully soon!).

although hopefully this isn't a sign that grok 4.8/5 isn't catching up to frontier because that would suck in cursor.

[Matthew Shen @flyme2bluemoon]

this is actually huge! one of grok bot's only shortcomings compared to dots was that grok 4.7 is just shy of frontier but this solves that!

grok bot was already leagues ahead and now they've just pulled ahead even more. the only remaining advantage oai has is computer use.

[Matthew Shen @flyme2bluemoon]

@catalinmpit I wouldn't recommend it because you might get even more addicted to vibe coding and it might ruin your sleep schedule (just one more prompt, I whisper to myself)

[Matthew Shen @flyme2bluemoon]

@venturetwins I think it's just @ThePrimeagen and @hankgreen holding down the fort out there...

[Matthew Shen @flyme2bluemoon]

@dhruvalgolakiya @poteto @thsottiaux i am sat, 1 front row ticket please 🎟️
🍿🍿

[Matthew Shen @flyme2bluemoon]

@poteto grok bot has quickly become one of my favourite products to use, and one of the most charming. it has also improved so quickly. it's now faster, talks better, and is more capable. there are still some kinks to iron out but i'm excited to see what the team cooks up next!

[Matthew Shen @flyme2bluemoon]

is grok @bot agi?

[Matthew Shen @flyme2bluemoon]

@AlexOnchain you might not believe me when i tell you this, but this place is real and it's one of the best cities in the world

[Matthew Shen @flyme2bluemoon]

it seems that the subagents sometimes don't report which model they are using. in this case the parent thread was gpt-6 astra. for one subagent it shows gpt-6 astra while for another subagent it shows not reported (although I'm pretty sure it's gpt-6.1 sol since that's what I asked for).

also, the main thread doesn't seem to wake up when a subagent finishes, if the main thread finishes first and enters the waiting state.

[Matthew Shen @flyme2bluemoon]

@AritraDas262577 @theo it was fixed here: https://t.co/1we7jUoDNX

ran into this yesterday and it already fixed today!

[Matthew Shen @flyme2bluemoon]

X is truly a special place

[Sundar Pichai @sundarpichai]

@steipete Ack, will follow up

[Matthew Shen @flyme2bluemoon]

@blueemi99 i find them helpful as a way to keep the main context window clean and allow for parallel exploration which results in more diverse outputs. they're also quite useful for getting agents to unblock themselves since they exhaust a larger search space.

they are rough on usage tho.

[Matthew Shen @flyme2bluemoon]

@emollick the trickle-down theory of ai

[Matthew Shen @flyme2bluemoon]

@sama we're surprised a company named Anthropic is anthropomorphizing its ai...

maybe we should have seen this one coming

[Matthew Shen @flyme2bluemoon]

as it turns out, competition is really good for the consumer. however, the flip side of this is that my wallet is now being pulled in a game of tug-of-war between openai, anthropic, and spacexai...

[Matthew Shen @flyme2bluemoon]

it really feels like we're in a golden age of ai models. both frontier labs have very good models that feel great to use with fairly generous usage limits, even on the $20 tier subscriptions (ChatGPT Plus/Claude Pro).

[Ben Davis @davis7]

My vid on Opus 5.5 and GPT 6.1 Sol:

- Sol 6.1 is miles better than 6.0, it's worth trying out & the price is nuts
- Opus 5.5 is the best Claude model in a very long time