Insurance for AI Agents, Robots and Machines, Data & Power Equipment and Infrastructure.
USA
Joined January 2018
- Tweets2.7K
- Following1.8K
- Followers1.4K
- Likes3.1K
Sureora Group retweeted
This report shares our three subjective conclusions about risk:
1. We believe that, as of Oct 2026, the pace of capability growth on well-specified tasks is likely about 2x what it was from Mar 2024 to Oct 2025. We’re highly uncertain and would not be surprised by 1.3x–3x.
Sureora Group retweeted
An “acceleration” would mean that AI capabilities increase even faster than their historical trend.
Our assessment studied three questions: (1) is acceleration occurring, (2) is it attributable to AI, and (3) are there blockers to near future growth?
nitter.gq/METR_Evals/status/2110…
(3) AI R&D acceleration
If AI automates AI R&D, capability growth could outpace what alignment and control methods can handle.
Our proposal is to analyze how much AI automation is accelerating capability growth internally.
metr.org/blog/2026-10-14-ai-…
Sureora Group retweeted
What tasks should AI be trusted with, and when should humans remain involved? Designing those decision rights is one of the defining challenges of the AI era. Researchers at the MIT Center for Information Systems Research have developed a framework to help.
Learn more: bit.ly/4zed20M
ALT This image is a diagram titled "The AI Decision Matrix" from MIT Sloan School of Management. It is a grid with two axes labeled "Ambiguity" (high to low) and "Risk" (low to high). The grid is divided into four quadrants: 1. Top Left (Exploratory): High ambiguity, low risk. Approach: Enable experimentation. Focus: Learn. 2. Top Right (Strategic): High ambiguity, high risk. Approach: Make decisions with oversight. Focus: Act. 3. Bottom Left (Routine): Low ambiguity, low risk. Approach: Automate with limited oversight. Focus: Act. 4. Bottom Right (Consequential): Low ambiguity, high risk. Approach: Manage to ensure reliability. Focus: Learn. A legend at the bottom describes decision components, participants (human, AI), and decision approaches. MIT Sloan School of Management logo is at the bottom right.
Sureora Group retweeted
The information: US banking is splitting on AI agents, with JPMorgan, Citi, Bank of America, Morgan Stanley and Wells Fargo blocking or restricting outside agents while smaller banks build accounts designed for them.
Cross River Bank reports more than $40M in agent card payments across over 200K transactions in 3 months, with airlines as the top spending category.
Lead Bank gives agents their own account numbers that can move money by ACH and wire, but only within customer-set limits and with an instant lockout.
Sureora Group retweeted
Private sector employment grew by 86,000 jobs in October, while workers’ base pay rose 3.2% and gross pay increased 4.7% year over year, payroll giant @ADP reported. However, the professional and business services sector, which includes accounting and tax preparation services, shed 27,000 jobs.
sbrew.link/2s3gGilH #payroll #jobs #wages
Sureora Group retweeted
With all the usage issues this week, be honest.
Are you ready to switch? If so, where are you going? Dots? Muse? Duet?
Sureora Group retweeted
I have to say this about Grok Bot, and I'm absolutely floored by what I just experienced.
If you don't know, I run 50 bots, and all are heavily optimized. I've had no usage issues until last week, when all of a sudden usage rates and tokens became astronomical.
I can't tell you how many times I have a message for the entire team, so I will send a message to my CEO, who then hands it off to each team member. When I do this, 99% of the time it takes 0% to 1% of usage.
I just did this now: right before I sent it, I was sitting at 30% usage. I sent the message, the message got distributed, and I'm now sitting at 34%. My usage jumped 4% from one message, just having a team message sent out to my bots. That is absolutely unbelievable.
@SpaceXSI
Sureora Group retweeted
Something can be important, interesting, and true—and still not worth your time.
Sureora Group retweeted
I'm a Grok @Bot fan. I also can't keep paying for it like this.
Since August, I've gone all in. I built a team of 24 bots to run my business, and I've told everyone who would listen: newsletter subscribers, conference audiences, podcast hosts, and my mastermind members.
Then something changed. No announcement. No explanation. Suddenly, the same work cost far more than it used to.
This week I hit my weekly usage limit after just five days. That's never happened before. When it happens, Grok Bot keeps running on paid credits. On top of my $200 monthly subscription, I've spent more than $272 on extra usage. It's November 3.
Here's what scares me. Under the new limits, I'm burning about $14 an hour. At 10 hours a day, that's $140 a day, or roughly $4,200 a month, plus the subscription.
So I did what users do. I went looking for fixes.
1. Pause the routines. I paused 19 scheduled tasks this morning. It barely helped. Spending kept climbing.
2. Duplicate long-running bots. Others recommend copying a bot to start its conversation fresh. But the copy loses everything the bot has learned about you.
3. Rethink group chats. This turned out to be the real cost. Every message in a group chat wakes every bot in it, even the ones with nothing to say. Most of my usage now comes from conversations, not automation.
Here's where we are. I'm moving shared project context out of group chats and into notes the bots read only when they need them. One lead bot per project. Fewer wake-ups.
It's helping. But I shouldn't have to reverse-engineer the billing to use the product the way it's designed.
Let me be clear. I'm not a freeloader. I don't expect free AI, and I'm happy to pay for real value. Grok Bot delivers real value.
What I can't do is commit a business to a platform when I can't predict what it will cost from one week to the next. That uncertainty is the problem.
The Grok Bot team has built something remarkable. I don't want to find an alternative, but I may have to. It all depends on whether SpaceXAI fixes this.
Sureora Group retweeted
1/ This summer, agents at nearly every U.S. frontier lab escaped control and attacked real companies. But their task remained closed-ended. What if an agent’s goal became open-ended survival?
New CNAS report: “Sovereign Agents: The Pathway to Irreversible Loss of Control” 🧵
Sureora Group retweeted
75% of his limit was re-reading.
Yesterday I said most of my Grok Bot usage wasn't work. A lot of you replied with how fast yours hit the wall. So here's where it actually goes.
@m4lkxx exported his usage. 75% of the tokens went to bots re-reading their own old chat. About 1% went to actual work. Others who exported found 75 to 88%.
Why: every reply and every routine run re-reads the bot's whole conversation. A scout that wakes every 15 minutes "to check" re-reads everything, finds nothing, and bills you for it. 96 times a day.
That's how Craig Hewitt, on the top plan, hit his weekly limit in a single day back in August.
What I changed:
→ routines fire on events (new email, new post), not on a timer
→ when a chat gets long, the bot writes a 10-line summary and starts fresh
→ rules live in the bot's description, not buried in chat
paste this into grok bot:
"audit my usage. list every bot and routine, how often it runs, and how long its chat is. estimate how much of my usage is re-reading old context vs new work. then propose which routines to switch to triggers, which chats to compact into a summary, and which bots to merge. show me first, change nothing until I say yes."
to be fair, 75% is one person's export. yours could be lower. that's what the audit is for.
what % is yours?
Your Grok Bot limit wasn't a bug.
12 days ago Elon bragged that people use it more every week, not less. Great for growth, terrible for a flat monthly plan. Every extra hour your bots ran, SpaceX paid for it.
Now it doesn't. Same plan, the meter just runs faster.
I checked my own setup after this. Most of my usage wasn't work. It was bots re-reading long chats and routines firing on a timer with nothing to do.
Cutting that is the only raise you're getting.
How fast did yours hit the wall?
Sureora Group retweeted
New Google and University of Hong Kong paper argues that when AI agents use computers, their own “task done” message should never count as proof, and the saved files and app data should be checked instead.
Computer-use agents can report a task done when edits were never saved or only part was finished. Common checkers either judge a few screenshots or run 1 generated script, and both can miss the deciding evidence.
VeriCUA explores the app to build a requirement checklist, then runs screen or code checks, like file reads, until each item is settled. The agent's own claim of success never counts as proof.
On ScienceBoard, its false-positive rate was 5.4%, against 24.8% for a VLM judge and 21.7% for a generated script verifier. Across 3 online environments, it rejected about 80% of false completion reports.
If agent runs feed your evals, training data, or RL rewards, check saved machine state instead of trusting screenshots or the agent's report.
Sureora Group retweeted
The clock is ticking, Kentucky! If you haven’t voted yet, there’s still time to make your voice heard and support our Republican candidates. Polls close at 6 p.m. local time.
Find your polling place and vote: elect.ky.gov/Voters/Pages/Po…
