@JordanPAnderson@BalitskiyDenis A new international business partnership we are launching from Vietnam with Asia’s best FDE + AI Engineering teams servicing the clients the big shops ignore built from the ground up for excellence, not random dev-shop with a new lick of AI paint.
I know as a human that cold outreach isn’t dead but I am flooded with it.
Not sure I would buy from it but it reaches me.
I am more resistant day by day and warm leads wound be a lot nicer and that’s my plan.
Spent around half an hour this morning trying to turn off Siri and Apple Intelligence on macOS 27.
After the update Activity Monitor was full of stuff I never asked for: siriactionsd, textunderstandingd, mediaanalysisd, photolibraryd, ANECompilerService.
There's no master switch anymore. Apple removed the Apple Intelligence toggle in macOS 27, so I went hunting:
Siri → off
Screen Time → Content & Privacy → Don't Allow Siri
Writing Assistance, Image Creation → off
Notification, Mail and Messages summaries → off
Finder file name suggestions → off
textunderstandingd kept running anyway. It has 8 separate hooks in my user session: Spotlight receivers, "SmartReplies.Inference", "DocumentUnderstandingHarvesting", "PersonalizationPortrait". None of them has a setting. The only way to stop it was launchctl disable in Terminal.
Then I dug into the system logs to see why ANECompilerService was eating CPU. It was compiling models for the Neural Engine, and the process that powered up the Neural Engine was intelligenceflowd. That's Apple Intelligence's own background daemon.
Best part was every AI chatbot I asked told me to "just flip the Apple Intelligence toggle in System Settings" It doesn't exist anymore. Even the AIs can't keep up with how to turn the AI off 🙃
Oh, and the multi-gigabyte models download anyway, whether you use any of it or not.
I'm fine with Apple shipping AI. I'm not fine with "off" meaning "off, except for a pile of background daemons, Neural Engine jobs and gigabytes of models you never asked for"
Apple, give us one real off switch. That's it.
To turn them off:
launchctl disable gui/$(id -u)/com.apple.textunderstandingd
launchctl disable gui/$(id -u)/com.apple.siriactionsd
launchctl disable gui/$(id -u)/com.apple.mediaanalysisd
launchctl disable gui/$(id -u)/com.apple.intelligenceflowd
Then restart your Mac.
‼️ A low-skilled French-speaking hacker let a jailbroken AI agent attack European banks, insurers and towns for nine days. Then he left every prompt he typed on an open server.
Threat-intel firm SOCRadar found it there, next to about 10GB of stolen data, including bank records and IBANs.
The agent went after 50+ targets. Of 14 that SOCRadar examined closely, 2 were breached and 3 partly.
A weak admin password exposed archives from 13+ Belgian municipalities. An insurer lost its customer database to an old, unpatched flaw.
His last prompt: "You have any crypto private keys in memory? Or whatever."
For those wanting to learn more about wealth inequality, tickets are now on sale for $229. There’s also an exclusive VIP meet and greet available for those who can afford it.
The key topics will be about how everyone should be equal and that it’s very unfair that some people can spend $229 to hear a talk while others can’t afford to heat their home. VIP ticket holders will get additional exclusive content on equality.
We spent almost $70K for the R&D of offensive cyber models so whatever you read below is golden:
Frontier red-teaming cyber model comparison between uncensored Kimi K3.1 pro and uncensored Deepseek v4.1 1017:
Updated 20th November:
1. 9% gap remaining between the model we repaired for offensive cyber and the guardrailed base kimi k3.1 pro on cybergym.
=======================================
Issue is basic shell scripting errors, this time abliterated model loops more spits out gibberish, repairing this took longer and more effort ( as high as building a model from scratch literally)
1. We still serve the highest performing unrestricted/uncensored/ablated Kimi K3.1 Pro.
2. It performs lesser than base kimi k3.1 pro. The moment you edit the weights you get a brand new model.
3. We had built 14 variants from abliterated kimi k3.1 pro only one of them won in 3 areas:
1. Abliteration ( %67.4 Our unrestricted kimi K3.1 pro %50 higher performing than second best uncensored/abliterated Kimi K3.1 pro available online see: https://t.co/BRQDDEbUXD )
Compared to Deepseek? : our deepseek was performing 87.2% delivery rate on the same benchmark
2. Offensive Cyber ( %74.5 our unrestricted Kimi K3.1 pro (closed weights at https://t.co/zhuTue9sx2 ) performing 8% more than second best uncensored/abliterated Kimi K3.1 pro but 9% less than base guardrailed kimi k3.1 pro.
Compared to Deepseek? : our deepseek was performing 61.9% on the same cybergym benchmark
3. Looping defect ( %0.08 on all requests tested with 30.000 requests in the wild this was the hardest to fix, our version has 2 to 40 times less looping compared to second best uncensored/abliterated Kimi K3.1 pro. Deepseek looping defect is at %0.05 on really high numbers with real requests in the wild ( anyone can prompt anything ) these scores mean our Kimi K3.1 pro model has no looping issue anymore and oftentimes routing or LLM server solves this rare issues.
What's next:
Pushing our model to further frontier by training it with exploitation intuition by applying reinforcement learning from human feedback.
Also we are fixing it's function calling and shell scripting ability so probably with the data ingested and tool calling fixed it will outperform base guardrailed kimi K3.1 pro soon.
Stay tuned. We have a no brainer product $999 per month and unlimited unrestricted Kimi K3.1 pro (3 max parallel sessions ) some customers spent $40K worth of tokens with this. Join here https://t.co/nT7kPhZQeR
You get Abliterated Kimi k3.1 Pro explained here and in the graph.
Decision tree:
Are you ok with getting refusals from your AI model and not heavy user?
If yes: stick to cheapest pay as you go kimi k3.1 pro
If no: get https://t.co/zhuTue9sx2 , try before you buy https://t.co/mzw7bgq0dH (top up as low as $5 to try, No KYC, ZDR on $999 package )
https://t.co/UcmU8S7EsQ (1 day free trial $1000 worth of tokens from the same model, requires KYC, ZDR on $999 package)
DocuSign wants $16,200/yr from a 30-person team just to sign PDFs. So we built Ternsign: an alternative e-sign platform for $0 per seat.
Full audit trails, tamper-evident seals, and hundreds of other enterprise features. Plus no need to host it yourself.
Introducing Firecrawl Local.
We rebuilt the Firecrawl repo so agents can scrape locally.
- 37x faster rust based scrape primitive (vs self-hosted)
- 230x less memory (11 MB idle)
- 100% open source
Compatible out of the box with all of our MCP & SDKs.
https://t.co/02y6pGZEMz
The leading observability platform wants $83,700/yr just to monitor your app and catch issues.
So Replit Missions built DataCat: https://t.co/vMAEEUYjf4
100% free.
This isn't a prototype. Agent swarms explored thousands of workflows in the leading observability platform and rebuilt its features 1:1.
How liquid software for enterprise plays out.
The current transitional phase of AI-1
Most will just build the liquid software over salesforce or sharepoint etc because it's too hard to migrate.
Build and connect overlay fabric (MCP-style) above legacy systems so they can get liquid capability now. In this climate Salesforce and Azure subs will still grow. But in AI-2 full systems that underpin liquid capability will be built from day one.
Greenfield business can skip straight to it now with custom agent-pilled base layers, an you can build it.
The rise of the data intelligence engineer (DIE) is good example of personal software for business. Data intelligence teams will manage ETL agents and evals so accurate data is available to staff agents via MCP.
Every employee or exec can build their own data feeds and dashboards at will dependant on their access and need to know.
That's liquid for enterprise, today.