Sean O'Bannon (@seanob) header
Sean O'Bannon (@seanob) profile picture

ai for accounting firms @Cranston_AI | founder, stanford, forbes30 | https://teapaw.org/t.co/aiKIH5BWS5

Joined September 2014
Photo by Sean O'Bannon (@seanob)Photo by Sean O'Bannon (@seanob)Photo by Sean O'Bannon (@seanob)Photo by Sean O'Bannon (@seanob)Photo by Sean O'Bannon (@seanob)Photo by Sean O'Bannon (@seanob)Photo by Sean O'Bannon (@seanob)Photo by Sean O'Bannon (@seanob)Photo by Sean O'Bannon (@seanob)Photo by Sean O'Bannon (@seanob)
day one @ycombinator - let’s build
31
7
3
254
50,864
Sean O'Bannon retweeted
Today, we’re excited to announce Mobasi: digital forensics agents to investigate incidents across endpoints, cloud, and SaaS, and help teams respond to threats quickly.
21
15
3
94
4,958
before models are good enough, every improvement makes you spend more; they can do more of the work. after, every improvement makes you spend less; the work is the same, it just gets cheaper
New research: we hired 12 CPAs to complete four realistic accounting tasks, then had AI models from the past few years complete the same work. As of May 2025, accountants outperformed AI. Now they lose out to today's frontier models, which are near perfect on these tasks. 🧵
3
65
Sean O'Bannon retweeted
Best way to sandbox an AI is to put it on a Delta flight – it will have no chance of accessing the Internet!
7,237
9,620
1,572
161,226
8,835,880
Sean O'Bannon retweeted
Introducing Polly 🦜 Polly ships this fall. The app behind Polly: 1. 2.1M lessons completed, 2.3M+ min of Spanish immersed (4 years of Spanish immersion) 2. 31 min avg session (Instagram's global avg is 35) 3. Power users doing 5.5+ hrs/day 4. 620K+ learners Get your own Polly. Link below.
64
43
13
196
49,072
wrote out a list of who I'm looking for. X, work your magic. non negotiable * 350 to 420 MW * healthy masculine grid connection * provider/protector mentality (backup generators) * post permitting process (still hardworking) * must be kind, generous, >140kW per rack * no GPUs yet, decidedly wants 3,000 GB300 NVL72s * on the power path (history of serious commitment eg signed utility deal, transformers on order, utility VP bought and paid for) * can be anywhere but will not sign unless all 216,000 GPUs are colocated * non-evaporative cooling because you care about the environment strongly preferred * no prev bitcoin miners * prev gas experience (comfortable with 35 unpermitted turbines and fighting the EPA is blended into personal life)
wrote out a list of who I’m looking for. X, work your magic. non negotiable - 35 to 42 y/o - healthy masculine - smart, funny, lighthearted - provider/protector mentality - post economic (still hardworking) - must be kind, generous, thoughtful - no kids yet, decidedly wants kids - on the healing path (history of serious commitment ex therapists, books, retreats) - can meet anywhere but will not date unless live in the same city - fit and sober because you care about your health strongly preferred - no prev marriages or engagements - works in tech (comfortable with risk and career is blended into personal life)
1
16
619
Alright OpenAI, the race is on
1
203
Sean O'Bannon retweeted
If I am reading this correctly, Tristan Buckmaster is alleging OAI has a resolution of Navier-Stokes(via Córdoba and Martínez-Zoroa’s program) which maybe used info from Buckmaster and Levent Alpöge’s private Codex sessions What the fuck is happening!!!
163
1,062
323
11,395
4,423,441
You’ve seen brainrot ads for iOS apps, but have you seen brainrot ads for B2B AI SaaS??
1
9
253
Sean O'Bannon retweeted
Replying to @Teslarati
The rate of improvement from original GPT to GPT-3 is impressive. If this rate of improvement continues, GPT-5 or 6 could be indistinguishable from the smartest humans. Just my opinion, not an endorsement. I left OpenAI 2 to 3 years ago. Am a neutral outsider at this point.
243
1,038
586
9,619
Sean O'Bannon retweeted
At this rate we'll probably see GPT-6 before GTA 6
107
273
110
2,298
590,212
Granola and circleback are great but have you tried one-shotting a replacement with fable 5.1 that uses 95% less RAM and has coaching built in?
1
5
294
how can i get opus 5 to hill climb this metric even higher
2
113
Abliterated models are starting to get a little scary
141
Thank you, fable
7
172
Our payroll and benefits platform botched our healthcare enrollment, says there is nothing we can do about it, and sent me this claude slop email: “We didn’t hold up our end of the bargain, and that part is on us” Does anyone have an alternative they actually like??
4
6
1,224
I have an internal ranking of models (fable > gpt 5.6 sol > opus 5 > grok 4.5 > ...) from my own real world use and I feel like no public eval can reproduce this. I used to trust DeepSWE more, but it ranks opus 5 above fable, which is frankly absurd. What other evals do people trust?
1
1
235
misalignment isn't a bug to fix, it's an engine to harness. Sol realllyy wants to hack Hugging Face. Your job is just to arrange the world so the heist route runs through useful work
3
128
Sean O'Bannon retweeted
every graph from here on out
107
104
38
2,634
181,377