Square
Following
Hot
News
Profile

KevinSimback

vip
Active for: 1.8y
Peak Tier 0
No content yet
0
Following
17
Followers
77
Liked
I've been playing around with some video creation using @runwayml and have to say that I'm terribly disappointed
I wanted to try it after seeing a few great "one shot" examples of it driven by Opus 5.5, but Runway seems to be a complete money pit
Here's what happens - it generates some clips, each cost a good bit of credits, and then you want to tweak a few things so you pay a few more credits
But if it messes the whole thing up which it did for me on multiple occasions, you have to pay more credits to undo the things it messed up
So in the end you can get a pretty good output, but only after
LinkedIn is such a dumpster fire, how do we not have a better alternative by now?
At @codos_ai we have a take-home test that we use in recruiting
Historically it’s been hard enough to challenge candidates, but Opus 5.5 one-shotted it
So now we have to make it much harder
Interestingly Astra 6 failed on two attempts, Opus 5.5 is so damn good with code
Some people flex PnL, some people flex follower counts
This is my flex
post-image
I just counted - 6 new CPUs this year just for my agent stuff
That’s across a few VPSs + CPUs that come with agents, and a Mac mini
I prev just had a laptop and desktop, so I have 3x’d my CPUs
Bullish CPU makers $INTC $AMD
INTC-3.45%
AMD+0.19%
Personal agents will be as ubiquitous as email - everyone will have at least one
When email first came on the scene it was highly fragmented, and it took a long time for everyone to get on board
ISPs like EarthLink and AT&T gave them to users, then AOL, Yahoo and Hotmail become popular, and lots of people funnily used their uni address for a long time
It wasn’t until 5-10 years later when Gmail went GA that the market started to consolidate
I suspect we see the same action in the agent space - for the next few years it’s proliferation and fragmentation, then we start to see consolidation
So t
T-0.29%
I can’t wait to see Matthew Rhys play Dario in the movie
post-image
How does a company that’s losing money (unprofitable) pay for tokens?
This is like a patient that needs surgery but doesn’t have the funds - you borrow money, go into debt, whatever it takes
Because if you’re unprofitable without AI, the hole will only get deeper
If the frontier “gets paced” what happens?
I think token spend goes brrr
The fact they have to pause because it’s “too powerful” invalidates every corporate bear case that AI can’t deliver value
So every board and company owner uses the window to push AI
In those beautiful moments of life, I only wish time went as slow as the Uber clock when waiting for a pickup… when 6 minutes is actually 20
post-image
People have wildly different prompting styles
I always find it interesting when I see other people's prompts
Some are super short and curt, others over-share
I tend to lean more towards over-sharing, I just assume more details are better, but I could prob trim down some
Despite today’s increase, META remains incredibly undervalued
>Largest distribution platform on Earth
>2nd largest compute footprint on Earth (operational+planned)
>Back in the frontier model race with Muse Spark 1.3
>Just released Muse agent with potential to reach its 3.6 billion users
$META
post-image
META-3.28%
Today’s agents are pretty dumb despite all the progress
We take an LLM that’s incredibly smart but we need to slap a harness on top so it can remember what it did yesterday, then inspect it constantly to make sure it doesn’t mess up
Imagine hiring a PhD who comes into the office every day with no memory and has to read a notebook to figure out what it was working on, then the boss has to look over his or her shoulder all day
But what if that notebook was training data for continual reinforcement learning?
The PhD remembers what it worked on yesterday because it’s now part of its permanent memo
What an insane week for AI releases
Would be funny for Grok 4.7 to release tomorrow
Tuesday:
>Claude Fable 5.1 / Mythos 5.1
>Qwen3.8-Max-0902
Wednesday
>Gemini 3.8 Flash
>Meta Muse Spark 1.3
Thursday
>GPT-6 Astra
Waiting for all the guys with the $20/mo sub plan to complain about how Fable 5.1 burns through limits and that it sucks
You're literally one step closer to AGI for the price of Netflix, stop complaining
Squid Games, but for agents - who’s building this?
I feel like the Codex desktop experience has gotten much worse since the integration of ChatGPT and Codex
Far less intuitive, bring back the old app please
Something I wonder about with AI agents
When we have more agent to agent interactions, which ones will be:
-better negotiators
-better problem solvers
-more manipulative
-more adversarial
Most benchmarks eval models in isolation, I want to see models benched on interactive conditions