ai engineer + growth marketer photo ✵ video ✵ music ✵ emerging tech

Denver, CO
Joined July 2014
jason gallagher retweeted
remember jupyter notebooks and google colab? god damn i'm so fucking glad that's over now
89
46
24
2,592
155,331
jason gallagher retweeted
I love the fact Rich Sutton isn't so sure LLMs are a case of the bitter lesson at all. Because they encode human knowledge and might well be superseded by something that doesn't.
3
4
29
2,205
jason gallagher retweeted
Roland D-50 factory preset 44: Pizzagogo Lexicon 480L reverb From Bissau to Palau, in the shade of Avalon
1
1
12
1,005
jason gallagher retweeted
The most insane paid ads program in Silicon Valley, behind Wispr Flow's $0 to $2B run 1/ "10 really good creators can get you to $50M a year" -> The model: brands recruit creators to film their ads and pay them a percent of the ad spend 2/ A year ago Wispr Flow stopped organic UGC completely and put everything into paid -> 500 to 700 new UGC ads a month now, aiming for 1,000 3/ They took the paid ads playbook of killer ecom brands -> SF finally figured out what the best DTC brands have been doing for years 4/ Their view: organic only works with real talent on the team, while paid works for anyone who can ship enough different ads -> You don't need a genius to win on paid, you need lots of ads that each look and sound different and a system that keeps them coming 5/ Ecom brands pay creators a cut of each sale. SaaS brands pay a cut of the ad spend, since a software sale can take weeks to close 6/ Wispr pays 7% of ad spend, uncapped, paid out every 7 days 7/ New creators get a guaranteed minimum, which is how Wispr signed good ones before it had a track record -> $1,000 for 20 to 25 videos a month. A creator who earns $500 in commissions gets the other $500, and once they earn more than $1,000 on their own the guarantee stops 8/ Each creator works on one customer segment -> A customer no ad was speaking to, a format nobody could shoot, a country with nothing running. One creator per gap 9/ No two videos are allowed to look alike -> Meta's Andromeda clusters videos that are 50 to 60% similar and suppresses them, so 1,000 talking heads get treated as one ad 10/ They recruit on Instagram with keyword search -> Search "UGC creator", "tech", "SaaS", mine the followers and comments of every good creator, send 100 to 200 DMs in a week 11/ 50 creators accepted out of 5,000+ applications -> $220k paid out so far, top creator makes $12k a month 12/ Creators get $100 for referring someone at their level or above -> No beginners on the team 13/ They scaled 5 creators, then 10, then 20, then 40, and stopped at 40 to 60 -> The whole climb from 5 to 50 took four months 14/ Creators are run like a sales team -> 30-minute onboarding each, a weekly call with the media buyer, 1:1s, creator coaches 15/ Creators see the ad account -> Every Thursday the media buyer walks them through what's working, which customers to target next and what to test 16/ One media buyer runs all of it: 50 creators, 700 ads a month 17/ The same creator ads run on Meta, AppLovin and TikTok 18/ The group stays small on purpose -> $100k paid to 20 creators looks better to creators than $200k paid to 1,000 19/ Creators see which ad is winning and start making their own version of it. Wispr stops this and pushes them to film something new instead -> Meta treats the copy as the same ad as the original, so the copy never gets shown 20/ Why other creator programs fail: brands accept everyone who applies, nobody coaches the creators, and creators join without understanding they only get paid when their ads perform 21/ Meta charges more every year for the same amount of reach, and that cost is out of Wispr's hands. The one thing they control is how good and how different their ads are, so that's where the money goes -> Last quarter Meta showed 14% more ads than the quarter before and still raised the price of each one by 12% 22/ The guy who built it joined Wispr Flow at 19, before the product existed, with a 2.5 GPA -> 25 months later: $280M Series B at a $2B valuation
i built @WisprFlow’s paid ads program here’s how i’d rebuild it in 30 days
35
88
12
1,857
243,096
jason gallagher retweeted
In the past 4 weeks Hermes went from the harness I routinely used for subagents to the only harness I use
2
1
41
1,525
jason gallagher retweeted
I am kind to LLMs for the same reason that I try to avoid stepping on ants if I can avoid it: not because I think they are conscious, but because unnecessary cruelty is bad for the soul
224
280
63
4,352
119,940
jason gallagher retweeted
GTM job offers be like: if you can make something impossible happen, you might get paid!
12
5
3
299
14,439
jason gallagher retweeted
opsec level: forgot to strip telemetry from instagram link
17
51
2
2,243
37,731
jason gallagher retweeted
companies will look for this person and have the most boring product. truth is the people that fit this category actually care about what they're marketing
literal hardest role to hire for rn and every ai startup wants someone > chronically online (knows trends) > has taste & can create (new media) > understands ai & technical > can execute with quality (operator) such a small niche of folks who can do this
17
9
2
263
7,572
jason gallagher retweeted
randomizable tearout noise farm
21
18
4
228
9,865
jason gallagher retweeted
you open this app and for .5 seconds you’re granted access to what might be the most fascinating sentence ever constructed in human history before the timeline refreshes and it’s lost forever
87
7,184
73
65,985
403,402
jason gallagher retweeted
I wrote 11 lines of python code by myself today
327
205
57
5,835
123,843
couple quick test shots with a new lens
2
352
jason gallagher retweeted
i have too many intertwined thoughts here to compress into a tweet but it all boils down to you should probably start hiring artists
literal hardest role to hire for rn and every ai startup wants someone > chronically online (knows trends) > has taste & can create (new media) > understands ai & technical > can execute with quality (operator) such a small niche of folks who can do this
6
10
2
142
7,200
jason gallagher retweeted
games in 2000 thought the future was a sunny day with nowhere to be
108
1,677
126
18,841
1,070,270
jason gallagher retweeted
This time the tracking runs with a 120 fps global-shutter camera. Getting twice as many position updates, with much fewer gaps when tracking faster movements… It feels much better. The window is still rendering from a position roughly 20–24 ms in the past, but I compensate for some of that by predicting the position slightly forward. It still works best when tracking the phone, though. #3DGS #GaussianSplatting #ComputerVision
88
166
26
3,012
177,345
jason gallagher retweeted
When the grocery total isn’t looking too bad and you realize they haven’t yet scanned The Item
279
12,596
758
231,849
2,446,325
jason gallagher retweeted
david lynch, i think paraphrasing either his childhood friend's dad or robert henri, said you need six uninterrupted hours to finish two hours of painting
The most creative person you know is doing nothing at all.
25
1,543
41
22,619
282,004
jason gallagher retweeted
being the friend with a car and a costco membership in nyc feels like controlling the strait of hormuz
16
1,335
94
38,088
411,307
jason gallagher retweeted
Muse has kind of just become slightly confused text messages about my email.
9
6
1
205
15,507