Kyle
@kylestratis.com
4.1K followers 2K following 480 posts
📊 Data, AI/ML, and Cloud consultant @ Stratis Data Labs 🚉 VPEng @ Sequel Institute 🤓 Startup advisor ✒️ Writing about MCP 🐍 Python ☁️ AWS 🧙‍♂️ Elixir 🐷Ham radio 🇺🇸↔🇬🇷 [Boston // Athens] Mostly tech things here
Posts Media Videos Starter Packs
My slides and code for @odsc.bsky.social West are finally done and submitted!

Time to pass out for 2-3 business days
You really believe that the researchers working on the next generations of these systems only have access to the mid-tier consumer-grade ones?
Reposted by Kyle
Over the past year, my lab has been working on fleshing out theory + applications of the Platonic Representation Hypothesis.

Today I want to share two new works on this topic:

Eliciting higher alignment: arxiv.org/abs/2510.02425
Unpaired learning of unified reps: arxiv.org/abs/2510.08492

1/9
I’m told the Early Release version of AI Agents with Model Context Protocol inspired some changes to the roots documentation for MCP. On the one hand: hell yeah, better docs! On the other hand: now I have to update the chapter again 🥲

But I’m excited so many people are reading it already!
Expand and refine Roots content in the Client Concepts doc by olaservo · Pull Request #1605 · modelcontextprotocol/modelcontextprotocol
Summary Add Design Philosophy section to Roots documentation to help understand the "why" and remove potential ambiguity around security claims. Motivation and Context The existing roots ...
github.com
This was shared a few months ago at one of the companies I work for, and the recent neuromorphic advances (there are several different strains of promising research happening simultaneously) are all exciting just on the efficiency side of things, not to mention all the other things
It's Early Release Day! AI Agents with MCP just had a major release and reorganization, with additional details added too all existing chapters, the client chapter (chapter 3) split into two chapters,
Reposted by Kyle
There's been a noticeable increase of AI/ML practitioners here lately, it seems like the antis have gotten bored or have blocked themselves into their own corner of Bluesky or the site has grown so much that they aren't as noticeable so people are feeling more comfortable coming here
MCP clients, of course! Unfortunately, there are fewer resources out there for building your own clients, and few clients in the wild that support all the features that the protocol provides for (the MCP Inspector has one of them!), but without clients, we wouldn't be able to use MCP at all.
Since the launch of Model Context Protocol, MCP servers have taken up much of the limelight. But for those of us building AI-powered intelligent applications, we want to be able to integrate servers, not just build them. But how would we do that?
Reposted by Kyle
I got introduced to @randyau.com's 'Data Cleaning IS Analysis, Not Grunt Work' post during the #dataBS Conf this week: www.counting-stuff.com/data-cleanin... . I just finished--it was a great read.

Here are some quotes and thoughts I'm walking away with 👇

1/9 #RStats
Data Cleaning IS Analysis, Not Grunt Work
Also, most data cleaning articles suck
www.counting-stuff.com
Reposted by Kyle
Supporting the #SomervilleMA man grabbed by guys in balaclavas even after they determined he was 'lawful'
www.universalhub.com/2025/support...
I really need to start also publishing to leaflet!
Fun fact, we *do* actually do something like this :)

Leaflet doesn't have this exact feature yet (backlinks to all related Bluesky discussions) — but similar: we parse the firehose for Bluesky posts w/ our quote param in their embed record…

…and show in a post where it's been quoted on Bluesky!
screenshot: For example, Leaflet is only interested in events concerning pub.leaflet.* records. However, Leaflet can also choose to listen to other events. If Leaflet wanted to add a feature that shows backlinks to Bluesky discussions of a Leaflet document, it would simply start tracking bsky.app.feed.post records too.
Reposted by Kyle
And new paper out: Pleias 1.0: the First Family of Language Models Trained on Fully Open Data

How we train an open everything model on a new pretraining environment with releasable data (Common Corpus) with an open source framework (Nanotron from HuggingFace).

www.sciencedirect.com/science/arti...
My great uncle (Nicholas Gage) is receiving one of Greece’s highest honors today, the Order of the Phoenix, for not only his work in bringing to light the struggles of the northern villages during the civil war, but his continued support of Greece over the years.
Greece to Honor Four Prominent Greek Americans in NY - GreekReporter.com
Kyriakos Mitsotakis will present Greece's prestigious Order of the Phoenix to four notable Greek Americans in New York on September 24
greekreporter.com
while following sporadic updates and commentary on the latest news in data and AI, join as many as 18 others in subscribing here:
The Signal Path
Cutting through the AI noise to find what actually matters for engineers, data practitioners, and the businesses that depend on them.
thesignalpath.xyz
I see why people let social media followers/subscribers get to their head: thanks to my appearance last week on the @realpython.com podcast, subscribers to The Signal Path, my newsletter, have more than doubled!

If you want to see my slow descent into insanity
Reposted by Kyle
As part of WIRED's Politics Issue, read @stevenlevy.bsky.social
on how the vibrant, outsider Silicon Valley he discovered as a young reporter decades ago, has changed utterly, to become a place bowing before Trump

With incredible art from @coldwarsteve.bsky.social

www.wired.com/story/silico...
I Thought I Knew Silicon Valley. I Was Wrong
Tech got what it wanted by electing Trump. A year later, it looks more like a suicide pact.
www.wired.com