
AJ Asver
3.1K posts

AJ Asver
@_aj
Building @ParchaHQ. Prev Product @compound, @brexhq, @coinbase, @google. Alum @ycombinator, @uniofoxford. Dad of twins, in-person maximalist, Coda Yoda.
San Francisco Katılım Ocak 2008
1.3K Takip Edilen2.6K Takipçiler
Sabitlenmiş Tweet

Grep just achieved SOTA on the three major deep research benchmarks, beating Perplexity, Google, Nvidia, OpenAI, and Anthropic.
We're a two-person founding team.
Grep AI@grepdotai
English
AJ Asver retweetledi
AJ Asver retweetledi
AJ Asver retweetledi
AJ Asver retweetledi
AJ Asver retweetledi
AJ Asver retweetledi
AJ Asver retweetledi

The Bank of England just put $1.8T in private credit under a stress test with 40+ firms including Apollo, Blackstone, KKR and every other major UK bank.
Deputy Governor Breeden warned today of a possible "private credit crunch akin to the banking sector."
But the real story is the regulatory contradiction underneath.

English
AJ Asver retweetledi
AJ Asver retweetledi
AJ Asver retweetledi
AJ Asver retweetledi
AJ Asver retweetledi

Nearly half of all US AI data centres planned for 2026 have been delayed or canceled.
12 GW planned. 5 GW under construction. The other 7 GW? Stalled by a bottleneck nobody talks about: high-voltage transformers.
Lead times went from 12-18 months pre-2020 to 36-48 months today. Many are manufactured in China. Subject to tariffs up to 145%.
English
AJ Asver retweetledi

Deepfake fraud in financial services hit a structural threshold in 2025.
Injection attacks on iOS biometric verification surged 1,151% in H2. Not presentation attacks (holding a photo to a camera). Injection attacks that bypass the camera entirely and feed synthetic video directly into the verification API.
Deloitte projects $40B in annual US losses by 2027.
English
AJ Asver retweetledi
AJ Asver retweetledi

YC just demoed 198 companies. Their strongest batch ever.
The standout trend isn't another AI wrapper. It's this: 38 companies - nearly 1 in 5 - are building for regulated industries.
Healthcare. Legal. Finance. Compliance.
The industries where getting it wrong isn't a bad UX. It's a lawsuit.
English
AJ Asver retweetledi

Based on your post, I’ve been trying something I call proactive compacting because I hate waiting for a compaction to complete mid-task. My Claude Code IDE checks for inactive sessions (ones I’m not interacting with right now) and compacts them if they are above 60%. Works nicely because I never get interrupted by compaction anymore.
English


We built GREP for people who do work where being wrong isn't an option.
Underwriting. Diligence. Compliance. Legal research.
If your output ends up in front of a board, a regulator, or a counterparty with their own lawyers - you need receipts.
Your first research request is on us: grep.ai
English

What does this mean if you're not a benchmarks person?
It means when GREP runs M&A diligence, it reads the actual SEC filings. When it screens for sanctions, it cross-references the actual registries. When it does underwriting research, it cites the actual primary sources.
Not summaries of summaries. The actual records.
English

People ask why we care so much about benchmarks @grepdotai.
Because our customers use Grep for serious work like corporate due diligence, Underwriting decisions, and RFP responses for contracts, where the decisions cost hundreds of millions.
When the stakes are that high, "pretty good" accuracy is a liability.
Here's what #1 actually means. 🧵

English





