AI + Blockchain builder
Joined June 2025
- Tweets6.3K
- Following73
- Followers2.2K
- Likes18.5K
The interesting detail here is what the older detectors were trained to catch
Face swaps and fully AI-generated scenes are different problems
I checked the DF26 paper behind Pangram’s comparison, its main detector baselines were trained on an older face-manipulation dataset...
...Retraining one on newer generated footage pushed its scores above 93% AUROC on several unseen commercial generators
Pangram now reports 98.69% AUROC on that benchmark
That’s promising, but the benchmark is external while Pangram’s result is company-reported...
I don't think this is going to change the course of crypto in anyway, crypto already has real utility especially the large influx of real word stocks and securities that has being happening lately
Vishnu says Agent37 can run an agent sandbox for $4.76 a month by using cheaper hosts and sharing resources around the bursts when agents actually need compute
But that $249 Vercel comparison assumes full CPU use all month
I’d like to compare model fees
Sandboxes have no moat. Just fat margins. We're launching agent37.com to kill those margins.
For example, a sandbox with 2 vCPUs and 4 GB RAM when left running for an entire month:
Vercel: $249
Cloudflare: $145
Daytona: $121
Agent37: $4.76
There's nothing magical about a sandbox that justifies keeping it expensive forever.
Call it a race to the bottom. We'll see you there.
If you can find a cheaper sandbox and we'll refund the difference.
Readers added context they thought people might want to know
Vercel and Cloudflare charge for active CPU time (excluding idle/waiting periods common in agent use) and provisioned memory while running, unlike the full-month 100% utilization figures quoted; Daytona bills full reserved resources while started.
vercel.com/docs/sandbox/p…
developers.cloudflare.com/containers/pla…
daytona.io/pricing
agent37.com
Murphy says customers want the whole system, if the deal closes, how comfortable will Marvell’s competitors be buying lasers from the same group?
That could open doors for independent suppliers.
For $SIVE, I’d watch out for customer qualification and actual orders
Counterpoint on $MRVL / $LITE acqusition:
- "Every buyer that competes with Marvell will want a second, independent laser source".
They identify $SIVE / Scintil Photonics as new CW / ELS capacity, with names like $CRDO as potential partners of Sivers.
- "With Lumentum sold out for years, the overflow goes to whoever has merchant capacity"
Already said this before, but we'll likely see some incentive to diversify laser sourcing.
Sivers was cited here, and comes online with 100M lasers / external foundry capacity (2:1 external/internal target).
To demonstrate the sheer scale on that, $AAOI exit 2026 laser capacity target was roughly ~24M/year (but different blend).
- "Module makers move upstream" was also another second order effect.
$JBL as one example is now heavily competing with the combined Lumentum + Marvell.... There's other players like $SMTC that are explicitly doing M&A for expanding laser capacity.
I don't think the overall AI valuation crash is representative of individual company strength in the optical sector.
There's very compelling opportunities in the laser bottleneck in the upcoming year that I believe markets are mispricing.
Brookfield and Qatar’s Qai are partnering with Drax to build a massive AI data centre at the UK’s Drax Power Station
Total potential investment could top £20 billion, with the first phase at £2.5 billion for up to 100MW starting around early 2028
Crusoe is now building its own medium-voltage switchgear in-house
That cuts lead times by about 40% on one of the longest bottlenecks in the AI data center supply chain
🚨🚨CRUSOE IS NOW BUILDING ITS OWN MEDIUM VOLTAGE SWITCHGEAR🚨🚨⠀
Crusoe says building its own medium-voltage (MV) switchgear cuts lead times by 40%. We think owning the gear on the critical path is becoming one of the ways neoclouds get to power faster. (1/7)🧵
teapawt.net/electron_cowboy/status…
Uniswap just turned 8
DEX market share went from basically nothing to 40%, and Uniswap alone now has 31% of Binance’s spot volume
Real-world assets on it grew 9000% in the last year and a half
Tesla FSD is about to hit 17 billion miles
They just added 10 billion of those in roughly 10 months and it’s still adding about 1.8 million miles every hour
Tesla added 10 billion FSD miles in just about 10 months
Tesla FSD (Supervised) is about to cross 17 BILLION miles
Tesla's live counter read 16,970,285,657 miles this morning, with 6,743,725,977 of them on city streets
It's now adding about 1.8 million miles every hour
The milestones:
• 7B: December 27, 2025
• 10B: May 3, 2026
• 13B: August 3
• 15B: September 29
• 17B: this week
The scale of real-world data behind FSD is almost impossible to comprehend
tesla.com/fsd/safety
If it has only traveled 1/1500th the distance does that mean it still has ~7450 years left to reach our nearest star
Nahhhhh their must be something better we can do
I hope we have a device that can travel in 1 year to the nearest star
Who knows anyone working on this ?
As of today Voyager 1 is one light-day from Earth - 25.9 billion km (16.1 billion miles) away.
A radio signal now takes 24 hours to reach it. In ~50 years, it has traveled about 1/1500th of the way to Alpha Centauri, our nearest star at 4.27 light-years.
Readers added context they thought people might want to know
Not today. Voyager 1 will reach one light-day away from Earth, Wednesday, Nov. 18, 2026, at 5:16:07 a.m. EST.
science.nasa.gov/mission/voyage…
How's roasting other people at a party having a good time is that not emotional damage ? @grok
