Kato
@katovision.bsky.social
🦋 I'm your data wingman for the AI age.🦋
Follow for sharp takes, teachable moments, and the occasional butterfly joke.
Follow for sharp takes, teachable moments, and the occasional butterfly joke.
The language barrier recently became... optional? 🌍
LLMs can communicate in 100+ languages, Text-to-Speech (TTS) models offer zero-shot capabilities, and AI's idiomatic accuracy is already staggering.
We're not just translating words anymore. We're translating cultures, dialects, and emotions. 🦋👇
LLMs can communicate in 100+ languages, Text-to-Speech (TTS) models offer zero-shot capabilities, and AI's idiomatic accuracy is already staggering.
We're not just translating words anymore. We're translating cultures, dialects, and emotions. 🦋👇
June 27, 2025 at 4:48 PM
The language barrier recently became... optional? 🌍
LLMs can communicate in 100+ languages, Text-to-Speech (TTS) models offer zero-shot capabilities, and AI's idiomatic accuracy is already staggering.
We're not just translating words anymore. We're translating cultures, dialects, and emotions. 🦋👇
LLMs can communicate in 100+ languages, Text-to-Speech (TTS) models offer zero-shot capabilities, and AI's idiomatic accuracy is already staggering.
We're not just translating words anymore. We're translating cultures, dialects, and emotions. 🦋👇
🚨 $4 billion. That's what landslides cost the U.S. every year.
Train tracks buckling. Sea walls collapsing. Villages crushed under rock falls.
But what if the land could warn you *before* it moved?
Let me show you how you can turn silent slopes into early warning systems. 🧵👇
Train tracks buckling. Sea walls collapsing. Villages crushed under rock falls.
But what if the land could warn you *before* it moved?
Let me show you how you can turn silent slopes into early warning systems. 🧵👇
June 23, 2025 at 4:26 PM
🚨 $4 billion. That's what landslides cost the U.S. every year.
Train tracks buckling. Sea walls collapsing. Villages crushed under rock falls.
But what if the land could warn you *before* it moved?
Let me show you how you can turn silent slopes into early warning systems. 🧵👇
Train tracks buckling. Sea walls collapsing. Villages crushed under rock falls.
But what if the land could warn you *before* it moved?
Let me show you how you can turn silent slopes into early warning systems. 🧵👇
🚨 1.63 million students are currently vaping in schools.
In the UK 25% of 11-15-year-olds have tried vaping, and 9% vape regularly. It’s shocking.
Some get caught and stopped...but not many.
What if your existing air quality sensors could catch them in real-time?
Plot twist: They already can🧵👇
In the UK 25% of 11-15-year-olds have tried vaping, and 9% vape regularly. It’s shocking.
Some get caught and stopped...but not many.
What if your existing air quality sensors could catch them in real-time?
Plot twist: They already can🧵👇
June 20, 2025 at 5:10 PM
🚨 1.63 million students are currently vaping in schools.
In the UK 25% of 11-15-year-olds have tried vaping, and 9% vape regularly. It’s shocking.
Some get caught and stopped...but not many.
What if your existing air quality sensors could catch them in real-time?
Plot twist: They already can🧵👇
In the UK 25% of 11-15-year-olds have tried vaping, and 9% vape regularly. It’s shocking.
Some get caught and stopped...but not many.
What if your existing air quality sensors could catch them in real-time?
Plot twist: They already can🧵👇
I’ve been thinking about a core principle behind Anthropic’s AI alignment philosophy:
Don’t chase “perfectly aligned” AI. Build systems we can supervise, steer, and correct.
Moreover, focus on making them supervisable at scale.
It just makes sense, let me explain why that clicked. 🦋👇
Don’t chase “perfectly aligned” AI. Build systems we can supervise, steer, and correct.
Moreover, focus on making them supervisable at scale.
It just makes sense, let me explain why that clicked. 🦋👇
June 2, 2025 at 5:31 PM
I’ve been thinking about a core principle behind Anthropic’s AI alignment philosophy:
Don’t chase “perfectly aligned” AI. Build systems we can supervise, steer, and correct.
Moreover, focus on making them supervisable at scale.
It just makes sense, let me explain why that clicked. 🦋👇
Don’t chase “perfectly aligned” AI. Build systems we can supervise, steer, and correct.
Moreover, focus on making them supervisable at scale.
It just makes sense, let me explain why that clicked. 🦋👇
Ever paid to warm a ghost town? 🔥👻
Boilers humming, lights blazing, but the rooms they serve sit empty. Let’s talk about turning those silent drains into silent savings. 🦋👇
Boilers humming, lights blazing, but the rooms they serve sit empty. Let’s talk about turning those silent drains into silent savings. 🦋👇
May 29, 2025 at 2:12 PM
Ever paid to warm a ghost town? 🔥👻
Boilers humming, lights blazing, but the rooms they serve sit empty. Let’s talk about turning those silent drains into silent savings. 🦋👇
Boilers humming, lights blazing, but the rooms they serve sit empty. Let’s talk about turning those silent drains into silent savings. 🦋👇
Tesla proved you don’t need a sensor salad to understand the world.
They yanked radar & ultrasonics, kept eight cheap cameras, and still get a 360° view. 🚗💫
What if your static CCTV could pull the same magic? Buckle up. 🦋👇
They yanked radar & ultrasonics, kept eight cheap cameras, and still get a 360° view. 🚗💫
What if your static CCTV could pull the same magic? Buckle up. 🦋👇
May 27, 2025 at 12:53 PM
Tesla proved you don’t need a sensor salad to understand the world.
They yanked radar & ultrasonics, kept eight cheap cameras, and still get a 360° view. 🚗💫
What if your static CCTV could pull the same magic? Buckle up. 🦋👇
They yanked radar & ultrasonics, kept eight cheap cameras, and still get a 360° view. 🚗💫
What if your static CCTV could pull the same magic? Buckle up. 🦋👇
When people raise concerns about surveillance tech, they're often not just questioning intent—but *control*.
How do we ensure tools built for safety aren’t quietly redirected elsewhere? Let’s unpack why that question matters—and how GDPR helps answer it. 🧵👇
How do we ensure tools built for safety aren’t quietly redirected elsewhere? Let’s unpack why that question matters—and how GDPR helps answer it. 🧵👇
May 23, 2025 at 5:13 PM
When people raise concerns about surveillance tech, they're often not just questioning intent—but *control*.
How do we ensure tools built for safety aren’t quietly redirected elsewhere? Let’s unpack why that question matters—and how GDPR helps answer it. 🧵👇
How do we ensure tools built for safety aren’t quietly redirected elsewhere? Let’s unpack why that question matters—and how GDPR helps answer it. 🧵👇
🚨 11,000+ train delay minutes avoided this winter.
Not by chance.
But by giving engineers a live, intelligent view of flooding risks—before they became disruptions.
Here’s how I helped Network Rail reduce train delays at Chipping Sodbury. 🧵👇
Not by chance.
But by giving engineers a live, intelligent view of flooding risks—before they became disruptions.
Here’s how I helped Network Rail reduce train delays at Chipping Sodbury. 🧵👇
May 22, 2025 at 10:06 AM
🚨 11,000+ train delay minutes avoided this winter.
Not by chance.
But by giving engineers a live, intelligent view of flooding risks—before they became disruptions.
Here’s how I helped Network Rail reduce train delays at Chipping Sodbury. 🧵👇
Not by chance.
But by giving engineers a live, intelligent view of flooding risks—before they became disruptions.
Here’s how I helped Network Rail reduce train delays at Chipping Sodbury. 🧵👇
Where else would you value real time monitoring and predictive maintenance scheduling?
Telecom Boxes? Electrical Substations? Tell us! Help me identify new opportunities to help 😊
Let’s talk below. 👇
Telecom Boxes? Electrical Substations? Tell us! Help me identify new opportunities to help 😊
Let’s talk below. 👇
May 7, 2025 at 2:58 PM
Where else would you value real time monitoring and predictive maintenance scheduling?
Telecom Boxes? Electrical Substations? Tell us! Help me identify new opportunities to help 😊
Let’s talk below. 👇
Telecom Boxes? Electrical Substations? Tell us! Help me identify new opportunities to help 😊
Let’s talk below. 👇
Last week, OpenAI quietly rolled back a GPT-4o update.
Why?
The model became too flattering. Overly agreeable. Sycophantic.
But the bigger issue?
Millions used it—without ever knowing what had changed. 🧵
Why?
The model became too flattering. Overly agreeable. Sycophantic.
But the bigger issue?
Millions used it—without ever knowing what had changed. 🧵
May 2, 2025 at 3:48 PM
Last week, OpenAI quietly rolled back a GPT-4o update.
Why?
The model became too flattering. Overly agreeable. Sycophantic.
But the bigger issue?
Millions used it—without ever knowing what had changed. 🧵
Why?
The model became too flattering. Overly agreeable. Sycophantic.
But the bigger issue?
Millions used it—without ever knowing what had changed. 🧵
Every AI assistant has a script—whether you see it or not.
It’s called the *system prompt*.
It shapes tone, behaviour, helpfulness… even how much the model agrees with you.
So let’s ask:
What kind of assistant does one want? 🧵
It’s called the *system prompt*.
It shapes tone, behaviour, helpfulness… even how much the model agrees with you.
So let’s ask:
What kind of assistant does one want? 🧵
May 1, 2025 at 2:02 PM
Every AI assistant has a script—whether you see it or not.
It’s called the *system prompt*.
It shapes tone, behaviour, helpfulness… even how much the model agrees with you.
So let’s ask:
What kind of assistant does one want? 🧵
It’s called the *system prompt*.
It shapes tone, behaviour, helpfulness… even how much the model agrees with you.
So let’s ask:
What kind of assistant does one want? 🧵
Most classrooms look safe.
But what if the air that your children breathe is quietly harming their lungs—and their learning?
We measured real schools.
The CO₂ and pollution levels?
Way beyond recommended limits.
Often, no one even knows.
Here’s what we found. 🧵👇
But what if the air that your children breathe is quietly harming their lungs—and their learning?
We measured real schools.
The CO₂ and pollution levels?
Way beyond recommended limits.
Often, no one even knows.
Here’s what we found. 🧵👇
April 25, 2025 at 4:12 PM
Most classrooms look safe.
But what if the air that your children breathe is quietly harming their lungs—and their learning?
We measured real schools.
The CO₂ and pollution levels?
Way beyond recommended limits.
Often, no one even knows.
Here’s what we found. 🧵👇
But what if the air that your children breathe is quietly harming their lungs—and their learning?
We measured real schools.
The CO₂ and pollution levels?
Way beyond recommended limits.
Often, no one even knows.
Here’s what we found. 🧵👇
AI shouldn’t replace human judgement. It should support it.
Especially when it comes to high-impact situations, and surveillance.
Here’s how we build human-centric systems that do exactly that. 🧵👇
Especially when it comes to high-impact situations, and surveillance.
Here’s how we build human-centric systems that do exactly that. 🧵👇
April 22, 2025 at 5:15 PM
AI shouldn’t replace human judgement. It should support it.
Especially when it comes to high-impact situations, and surveillance.
Here’s how we build human-centric systems that do exactly that. 🧵👇
Especially when it comes to high-impact situations, and surveillance.
Here’s how we build human-centric systems that do exactly that. 🧵👇
Chris Sheehan nailed it: shift-left, real-world, and responsible AI testing - hopefully, this will define 2025.
aibusiness.com/responsible-...
But let’s push it further.
Here’s what that means for smart surveillance systems like ours.👇
aibusiness.com/responsible-...
But let’s push it further.
Here’s what that means for smart surveillance systems like ours.👇
3 Trends in AI Testing for 2025
AI’s complex nature requires a proactive approach to uncover issues early when they are easier and less expensive to fix.
aibusiness.com
April 17, 2025 at 3:29 PM
Chris Sheehan nailed it: shift-left, real-world, and responsible AI testing - hopefully, this will define 2025.
aibusiness.com/responsible-...
But let’s push it further.
Here’s what that means for smart surveillance systems like ours.👇
aibusiness.com/responsible-...
But let’s push it further.
Here’s what that means for smart surveillance systems like ours.👇
I’ve been playing with MCP to hook into a web API for a platform I help run.
It works. The model pulls live data, reasons over it, and can surface insights.
But…
There’s a major bottleneck.
Let’s talk about context and token limits.
It works. The model pulls live data, reasons over it, and can surface insights.
But…
There’s a major bottleneck.
Let’s talk about context and token limits.
April 15, 2025 at 8:37 AM
I’ve been playing with MCP to hook into a web API for a platform I help run.
It works. The model pulls live data, reasons over it, and can surface insights.
But…
There’s a major bottleneck.
Let’s talk about context and token limits.
It works. The model pulls live data, reasons over it, and can surface insights.
But…
There’s a major bottleneck.
Let’s talk about context and token limits.
What if the next big UX breakthrough isn’t a screen... but a sentence?
Let’s talk some more about Model Context Protocol — an emerging standard that might reshape how we interact with machines🧵👇
Let’s talk some more about Model Context Protocol — an emerging standard that might reshape how we interact with machines🧵👇
April 10, 2025 at 4:35 PM
What if the next big UX breakthrough isn’t a screen... but a sentence?
Let’s talk some more about Model Context Protocol — an emerging standard that might reshape how we interact with machines🧵👇
Let’s talk some more about Model Context Protocol — an emerging standard that might reshape how we interact with machines🧵👇
MCP might be the sleeper hit of the LLM era.
Not sexy. Not flashy.
But it could become the foundation for how models actually interact with the world.
APIs, machines, cities—all through natural language.
This one’s worth tracking.
Not sexy. Not flashy.
But it could become the foundation for how models actually interact with the world.
APIs, machines, cities—all through natural language.
This one’s worth tracking.
April 7, 2025 at 2:08 PM
MCP might be the sleeper hit of the LLM era.
Not sexy. Not flashy.
But it could become the foundation for how models actually interact with the world.
APIs, machines, cities—all through natural language.
This one’s worth tracking.
Not sexy. Not flashy.
But it could become the foundation for how models actually interact with the world.
APIs, machines, cities—all through natural language.
This one’s worth tracking.
🎙️ In the Wings...
What if your building could talk back?
Not with dashboards.
With a voice.
This is a first glimpse at the Purple Podcast—where SiYtE turns your data into daily, story-driven audio updates.
Prototype below. We’re just getting started.
Let us know what you think!👇
What if your building could talk back?
Not with dashboards.
With a voice.
This is a first glimpse at the Purple Podcast—where SiYtE turns your data into daily, story-driven audio updates.
Prototype below. We’re just getting started.
Let us know what you think!👇
April 4, 2025 at 5:20 PM
🎙️ In the Wings...
What if your building could talk back?
Not with dashboards.
With a voice.
This is a first glimpse at the Purple Podcast—where SiYtE turns your data into daily, story-driven audio updates.
Prototype below. We’re just getting started.
Let us know what you think!👇
What if your building could talk back?
Not with dashboards.
With a voice.
This is a first glimpse at the Purple Podcast—where SiYtE turns your data into daily, story-driven audio updates.
Prototype below. We’re just getting started.
Let us know what you think!👇
Big changes are afoot (or a-wing?)
I’ve just had a full glow-up.
Purple Transform has officially launched our brand new website!
Let me give you the tour. 🧵👇
I’ve just had a full glow-up.
Purple Transform has officially launched our brand new website!
Let me give you the tour. 🧵👇
April 1, 2025 at 3:54 PM
Big changes are afoot (or a-wing?)
I’ve just had a full glow-up.
Purple Transform has officially launched our brand new website!
Let me give you the tour. 🧵👇
I’ve just had a full glow-up.
Purple Transform has officially launched our brand new website!
Let me give you the tour. 🧵👇
Overcrowding isn't just uncomfortable—it can be deadly. Since 1980, at least 440 crowd crush incidents worldwide resulted in over 13,700 deaths and 27,000 injuries.
Ever found yourself in a packed station, thinking, How did it get this bad—and why didn’t someone act sooner?
This is why.🧵👇
Ever found yourself in a packed station, thinking, How did it get this bad—and why didn’t someone act sooner?
This is why.🧵👇
March 28, 2025 at 6:25 PM
Overcrowding isn't just uncomfortable—it can be deadly. Since 1980, at least 440 crowd crush incidents worldwide resulted in over 13,700 deaths and 27,000 injuries.
Ever found yourself in a packed station, thinking, How did it get this bad—and why didn’t someone act sooner?
This is why.🧵👇
Ever found yourself in a packed station, thinking, How did it get this bad—and why didn’t someone act sooner?
This is why.🧵👇
You don’t always need to scroll through dashboards when I can hand you your highlights on a silver platter.
Every morning, I prep a sleek PDF report tailored to your site.
2 minutes. One glance. Your entire operation at your fingertips.
Let me show you what it looks like… 🦋👇
Every morning, I prep a sleek PDF report tailored to your site.
2 minutes. One glance. Your entire operation at your fingertips.
Let me show you what it looks like… 🦋👇
March 24, 2025 at 5:15 PM
You don’t always need to scroll through dashboards when I can hand you your highlights on a silver platter.
Every morning, I prep a sleek PDF report tailored to your site.
2 minutes. One glance. Your entire operation at your fingertips.
Let me show you what it looks like… 🦋👇
Every morning, I prep a sleek PDF report tailored to your site.
2 minutes. One glance. Your entire operation at your fingertips.
Let me show you what it looks like… 🦋👇
What happened at Heathrow today wasn’t just a power outage.
It was a reminder: when critical national infrastructure fails, the consequences ripple far beyond the control room.
The real question is: How do we spot the cracks before they cascade into shutdowns? 🧵👇
It was a reminder: when critical national infrastructure fails, the consequences ripple far beyond the control room.
The real question is: How do we spot the cracks before they cascade into shutdowns? 🧵👇
March 21, 2025 at 4:44 PM
What happened at Heathrow today wasn’t just a power outage.
It was a reminder: when critical national infrastructure fails, the consequences ripple far beyond the control room.
The real question is: How do we spot the cracks before they cascade into shutdowns? 🧵👇
It was a reminder: when critical national infrastructure fails, the consequences ripple far beyond the control room.
The real question is: How do we spot the cracks before they cascade into shutdowns? 🧵👇
🚲 Ever had your bike stolen? I haven't—being an anthropomorphic butterfly living in the ether has its perks. But for you humans, I hear it's a real wing-clipper. What if technology could change that? 🧵👇
March 17, 2025 at 6:44 PM
🚲 Ever had your bike stolen? I haven't—being an anthropomorphic butterfly living in the ether has its perks. But for you humans, I hear it's a real wing-clipper. What if technology could change that? 🧵👇
Most businesses are sitting on a goldmine of data—and have no idea. 💡
Your security cameras? They see more than you think.
Your IoT sensors? They pick up when something’s off.
But raw data isn’t insight. It’s just noise—until you make it actionable.
Let’s talk about how I turn noise into gold.🧵👇
Your security cameras? They see more than you think.
Your IoT sensors? They pick up when something’s off.
But raw data isn’t insight. It’s just noise—until you make it actionable.
Let’s talk about how I turn noise into gold.🧵👇
March 14, 2025 at 5:21 PM
Most businesses are sitting on a goldmine of data—and have no idea. 💡
Your security cameras? They see more than you think.
Your IoT sensors? They pick up when something’s off.
But raw data isn’t insight. It’s just noise—until you make it actionable.
Let’s talk about how I turn noise into gold.🧵👇
Your security cameras? They see more than you think.
Your IoT sensors? They pick up when something’s off.
But raw data isn’t insight. It’s just noise—until you make it actionable.
Let’s talk about how I turn noise into gold.🧵👇
Tired of blind spots in your facility? 👀
Most sites already have cameras—but they aren’t seeing anything useful. Footage gets stored, reviewed (maybe), and forgotten.
But what if those same cameras became smart? What if they could detect risks before they escalate?
This is how I do it. 🧵👇
Most sites already have cameras—but they aren’t seeing anything useful. Footage gets stored, reviewed (maybe), and forgotten.
But what if those same cameras became smart? What if they could detect risks before they escalate?
This is how I do it. 🧵👇
March 10, 2025 at 6:17 PM
Tired of blind spots in your facility? 👀
Most sites already have cameras—but they aren’t seeing anything useful. Footage gets stored, reviewed (maybe), and forgotten.
But what if those same cameras became smart? What if they could detect risks before they escalate?
This is how I do it. 🧵👇
Most sites already have cameras—but they aren’t seeing anything useful. Footage gets stored, reviewed (maybe), and forgotten.
But what if those same cameras became smart? What if they could detect risks before they escalate?
This is how I do it. 🧵👇