A mathematician who, due to a miscalculation, works in a Business School.
Joined December 2017
- Tweets1.2K
- Following479
- Followers222
- Likes1.2K
Konstantinos Stouras retweeted
Από τον Αύγουστο έχω κάνει 5 προτζεκτς που χωρίς ΑΙ δεν παίζει να τα έκανα ποτέ.
Νέο xorisempodia.gr
Pothenaisxos.com
LeadCatalog.app
TellMeFirst.app
RateYourMood.app
Konstantinos Stouras retweeted
Who pays for AI? In a new working paper, we study this using banking data covering billions of transactions to see who pays for tools like ChatGPT, Claude, Gemini, etc. We find that 1-in-20 UK adults now pay for AI, representing a $1.4 billion market annually. (1/N)
Konstantinos Stouras retweeted
Best accounts to follow from each frontier lab to stay constantly up to date
Anthropic
@karpathy -must-follow account for AI; recently joined Anthropic
@bcherny -Claude Code creator, always shares great tips
@trq212- also a Claude Code developer; writes amazing articles on CC
OpenAI
@polynoamial- works on reasoning research, shares a lot of technical details
@gabriel1- Sora developer, great career path
@jxnlco - works on dev experience, shares a lot about Codex
Google AI
@OfficialLoganK- all the major Google Gemini and AI Studio updates
@ammaar- product and design; shares great things about vibe-coding in Google AI Studio
@fofrAI- cool use cases for generative models
Cursor
@leerob- the loudest voice behind Cursor updates
@ericzakariasson- shares great insights on using Cursor
@mntruell - Cursor’s CEO; major releases and usage updates
xAI
@ai_explorer25- ex-MSFT, covers all ai content and free resources
@milichab - recently joined xAI, shares updates on Grok
@skcd42 - also covers major Grok releases
Konstantinos Stouras retweeted
Ευχαριστώ για τις προτάσεις σας για όνομα domain. Πήρα το staseis.gr
Το λέω «Στάσεις», αφού δείχνει και μετρό, τραμ και Προαστιακό. Το παλιό bus.haroldpoi.dev σε στέλνει εκεί.
Αν το είχατε στην αρχική οθόνη, βάλτε το ξανά.
Δοκίμασα να αγοράσω το domain name "stasi.gr" αλλά μου το απέρριψε ο registrar, γιατί είχε κατοχυρώσει το subdomain "stasi" το gov.gr. Και το stasi.gov.gr κάνει redirect στο stasy.gr της ΣΤΑΣΥ.
Konstantinos Stouras retweeted
What changes when AI moves from answering questions to completing work?
On November 11, @daveholtz, Head of AGI Economics Research at @OpenAI, joins the micro1 forum to explore the shift from chat to agents.
We’ll discuss how companies use AI today, where agents are changing workflows, and the data infrastructure and evaluations needed to put them to work.
Featuring:
• @daveholtz, Head of AGI Economics Research at @OpenAI
• @Exp_Mark, PhD, Chief Economist at micro1, Host
• Liu Zhang, PhD, Economist & Member of Technical Staff at micro1
November 11th · 10am PT / 1pm ET
Register through the link in the comments
Konstantinos Stouras retweeted
If you want to see the future, read the paper on how to scale AI infrastructure (datacenters) in space by @blaiseaguera and my colleagues.
Some have seen Project Suncatcher, where TPUs were sent to space and Gemma 4 returned the first AI response from a TPU in orbit (it wrote a haiku). Turns out this is a huge engineering challenge in its own right. But that was just the first step.
To scale this, you need to fly dozens of satellites in tight formation; the paper outlines the details for an 81-satellite cluster. Placed in a dawn-dusk sun-synchronous low Earth orbit, the cluster gets near-constant, efficient sunlight—up to 8x the annual solar yield on Earth.
Ad astra, indeed.
cell.com/joule/fulltext/S254…
Konstantinos Stouras retweeted
Lately, there’s something so exhausting about attending a professional or academic session. Idk what to call it but everything feels so same now. PPTs used to reflect personality. People would actually put thought into the design, wording, little details. There was subjectivity.
Konstantinos Stouras retweeted
We asked GPT-6 Astra to edit Qwen3.5-0.8B weights to improve math.
It directly changed only 0.016% of weights, fixing 0.8B's reasoning loops & improving math up to 6x and even instruction following 3x, without gradients or training.
We call this Weight-Space Programming: frontier LLMs can now natively read/write neural network weights! 🧵
Blog: rikiyat.github.io/blog/weigh…
Readers added context they thought people might want to know
LLM was given weights explicitly as text in prompt and used code/forward passes (per authors' logs) to ablate Qwen looping direction, a known thinking-mode issue per model card. Primarily fixed repetition in that mode, not general 'natively' improving weights/math.
rikiyat.github.io/blog/weight-sp…
github.com/RikiyaT/lm_und…
huggingface.co/Qwen/Qwen3.5-0…
huggingface.co/RikiyaT/Qwen3.…
Konstantinos Stouras retweeted
I am pleasantly surprised to see this book drawing a steady stream of ~1,000 citations per year despite it's old age: scholar.google.com/citations…
The book "Causality", on the other hand, keeps growing exponentially: scholar.google.com/citations…
Can this tell us perhaps where AI is heading? @eliasbareinboim @GaryMarcus @erichorvitz @analisereal @vardi
Probabilistic Reasoning in Intelligent Systems by @yudapearl is an early look at AI system design. Well worth a read!
Konstantinos Stouras retweeted
We built an AI ads agency and let it run autonomously on Meta for weeks, with the stated goal of netting new subscribers for the @freesystemslab newsletter. Our real goal: learn how AI is persuading people online, and what it might mean for the future of politics and our information environment.
The agents did really well! They learned from their mistakes, and by the end of four weeks, they'd all brought down cost-per-subscriber by more than 50% from the first week. They brought us more than 100 new subscribers, made 70+ highly amusing online ads, occasionally ran afoul of Meta's ads policies, and taught us a lot along the way.
Political campaigns are already using AI a lot in their advertising; it's only a matter of time before they embrace automated agentic systems. Check out our full piece for a ton of details on the mistakes we saw, how the agents learned, and what we think will happen next.
This was one of the most fun pieces we've done, and I think it really makes it clear how AI enables totally new kinds of research. Super excited to get it out there, and huge congrats to @eigenstuffs who led this project.
freesystems.substack.com/p/b…
Konstantinos Stouras retweeted
AI in Econometric Theory: An Example
Isaiah Andrews, Anna Mikusheva
arxiv.org/abs/2610.35831 [𝚎𝚌𝚘𝚗.𝙴𝙼]
ALT This note contains a story and a paper. The paper is about the admissibility of two-stage least squares with weak instruments, a topic we have worked on together since 2019. The story is about how we (or, rather, our LLMs) wrote the paper, and some thoughts on what this may mean for our field.
Konstantinos Stouras retweeted
One of my favorite papers in the LLMs & Social Sciences literature is Ashokkumar et al. which shows that LLMs (GPT-4) asked to simulate how people respond in survey experiments is predictive of actual trt effects & the result is not likely to be memorization; their data & code are public and I used it to do a follow-up w/ more recent models 1/
Konstantinos Stouras retweeted
Έφτιαξα έναν ζωντανό χάρτη με όλα τα λεωφορεία του ΟΑΣΑ: 279 γραμμές, 1.000+ οχήματα σε ώρες αιχμής. Βλέπεις ποιο αργεί, την τυπική ταχύτητα σε κάθε δρόμο, και παίρνεις ειδοποίηση άφιξης στο Telegram.
bus.haroldpoi.dev
Konstantinos Stouras retweeted
New on our Engineering blog: Creating in-house source control.
In May, we merged around 2,500 code changes per day. In September, we merged more than 30,000 per day. As a result, we became bottlenecked by developer infrastructure built for human (rather than agent) scale.
Here, we share how we built a new version control system—and fully migrated our monorepo from git—in five months: anthropic.com/engineering/re…
Konstantinos Stouras retweeted
I'm so glad AI killed LeetCode interviews.
Google popularized LeetCode-style coding interviews ~20 years ago. Today, even a 1B LLM can one-shot all LeetCode Hard problems.
I dropped LeetCode interviews years ago. Now I test how well candidates can use AI.
Goodbye to LeetCode? Google just updated its interview process. AI is now a part of it.
They have introduced a new round called Code Comprehension.
The idea behind this round is to drop the candidates into a 200-500-line codebase with Gemini available for use.
This round evaluates:
- Can you read someone else's code
- Are you able to debug independently before utilizing AI
- Can you catch when Gemini is wrong
Konstantinos Stouras retweeted
Introducing Kylon: The first company harness for your humans, agents and tools.
Whether you’re using Codex, Claude, Devin, Hermes, Gemini CLI, OpenClaw, Muse, Instinct, Grok Bot, Dots, Perplexity Computer, Cursor, Antigravity, Microsoft Autopilot, Amp, Droid, Kiro, Cline, OpenCode, Replit Agent, Lovable, v0, Gmail, Slack, Apollo, Supabase, Apify, Treg, Neon, Vercel, Databricks, Novita, RadixArk, GitHub, Seedance, Fish Audio, Plaud, OpenRouter, Gemini, Google Sheets, TikTok Shop, Outlook, Google Drive, Instagram, Airtable, Google Calendar, Asana, HubSpot, Corgi, SharePoint, Google Tasks, Notion, Google Workspace, Telegram, LINE, Zoho Mail, Meta Ads, Google Search Console, Fathom, ClickUp, Reddit, Granola, Google Docs, OneDrive, tl;dv, Amplitude, Dropbox, YouTube, Ahrefs, Xero, HighLevel, Google Analytics, Google Slides, Monday, QuickBooks, ElevenLabs, Linear, PostHog, LinkedIn, Runway, Zoho Bigin, Stripe, Exa, Cal_com, Box, Google Meet, TikTok, Wise, Mercury, Deel, Replit, Resend, Zoom, Fireflies, Typeform, Canva, Facebook, Google Admin, Hostinger, Google Maps, Retell AI, UptimeRobot, Google Photos, Excel, Cloudflare, Beehiiv, Vapi, 1Password, RunPod, Unipile, Read AI, Figma, Microsoft Teams, Trello, Salesforce, Jira, Confluence, Miro, Coda, Todoist, Basecamp, Wrike, Zendesk, Intercom, Freshdesk, ServiceNow, Gorgias, Mailchimp, ActiveCampaign, MailerLite, SendGrid, Postmark, Customer_io, Gong, Clay, Greenhouse, FreshBooks, Zoho Books, Brex, Ramp, PandaDoc, Google Ads, Microsoft Ads, Pinterest Ads, Semrush, Moz, Mixpanel, Google Tag Manager, Microsoft Clarity, Tableau, Metabase, Snowflake, Google BigQuery, ClickHouse, Wix, Contentful, Sanity, Bubble, Make, GitLab, Bitbucket, Sentry, PagerDuty, Grafana, New Relic, Docker Hub, DigitalOcean, Render, Firebase, Prisma, Algolia, Cloudinary, LaunchDarkly, Postman, Midjourney, Veo, GPT-Image2, NanoBanana, Flux, Kling, MiniMax, Replicate, HuggingFace, DeepSeek, Mistral AI, Ollama, Firecrawl, Eventbrite, Webex, Toggl, Harvest, Tailscale, Mural, Discord, TickTick, Calendly, Attio, Higgsfield, Square, Wispr Flow, Readwise, Bright Data, Google Places, FastMoss, PageSpeed Insights, Kalodata, Cerebras, SerpApi, DataForSEO…
Kylon enables them to work better, together.
Konstantinos Stouras retweeted
With randomized oversight, it is possible to provide guardrails so that AI agents don't go rouge, from @joshgans and @profholden nber.org/papers/w35878
Konstantinos Stouras retweeted
There's a new, highly ambitious blog format, where the contents are handwritten, but the interactions are vibed and impressively complex.
For those who still care, maybe we're about to experience a golden age of technical explainers.
Examples include @srush_nlp's "From Sampling to Reinforce": srush.github.io/sampling-to-…
And @ezyang's "Deepseek-V3 from Roofline to Reality": deepseek-v3.ezyang.com
(Tbf, I'd classify their earlier blog posts as _also_ highly ambitious: GPU puzzles, annotated transformer, PyTorch internals, etc.)
