Joined July 2015
- Tweets 12,763
- Following 353
- Followers 1,314
- Likes 13,488
849 Photos and videos
Pinned Tweet
Mar 3
Wrote a new note on file-first systems and agents as a query layer over source documents.
1
2
2
399
"condescendingly", lmao 😂😂
17
Jun 14
This is not accurate. The overflow caveat doesn't apply when a 64 bit type is in use. 1 trillion -> 100 trillion kobo -> 10^14, BIGINT holds ~9.22 * 10^18, a lot of headroom. Overflow becomes a concern in 32 bit INT, and in this case the max kobo stored is ~2 billion -> ~21M ngn.
They were storing balances in kobos. 😆
I asked, "If a customer has 1 trillion, how will you represent it in the database?"
That becomes 100 trillion kobos.
Then comes the real question: How are you handling overflow?
1
228
Jun 12
There’s this startup I saw months back, something around yard sales, can’t find their handle. Any one with info??
2
1
2
1,901
Jun 12
Fable just downgraded to Opus because I asked it to review potential attack surface and to validate an IDOR hyp I had, WOW!! I wonder how security folks are able to use this model.
74
Jun 11
Re safety alignment and how models like Fable are shipping with aggressive refusals, I can see blindspots attackers can utilise. An attacker could sprinkle bio weapon build prompts inside their malware, triggering LLM safety refusals and causing AI scanners to skip analysis.
20
May 28
Can’t read through the tl without getting fatigued. Maybe I need a break from this app
1
109
May 28
😂😂
May 28
I caught up with a friend who was an EM in Big Tech and decided take a sabbatical over the last year. He completely unplugged from tech during this period.
I was explaining everything that happened from the December launch of Opus, getting AI pilled, EMs writing code again, OpenClaw, Tokenmaxxing, and now the Token Budgeting and questions around the ROI of AI.
He calmly looked at me and said "maybe I should take another year off and let the dust settle a bit."
79
May 22
“Aghh, you for talk since oh”
1
186
May 4
So much compute unused. No wonder cursor went with them.
xAI’s GPU fleet is running at about 11% utilization, exposing how hard it is for AI labs to fully use expensive Nvidia hardware.
Read more in our AI Agenda newsletter: thein.fo/4cHRjWI
91
May 4
You can't be a senior eng and not know how bad or how to spot CWE-798
2
948
Apr 15
Anyone with info, help out here. Are these NG breaches and hacks really being taken seriously? I haven’t seen any posts from the affected organisations.
3
91
Apr 14
For newline feed, I prefer ctrl j to shift Enter. It feels more smooth
82
Apr 8
I need someone with context on this to help me out here: when we talk about AI security, isn’t the focus supposed to be on hardening the harnesses? The llms produce tokens and the harnesses executes this. If the llm say run ‘rm -rf ~/’, only the harness can execute this.
47
Apr 7
I'm curious about services that serve mostly Nigerian-based users and what architecture is mostly favoured, active-active or active-passive?
1
258