Kevin O'Connor

Partner at @ScOpVC, a VC firm specializing in pre-growth AI companies. Past: former CEO and founder of @DoubleClick, CEO and founder of @GraphiqHQ

[Kevin O'Connor @kjpoconnor]

@MattWalshBlog Let me know if you'd like to actually learn about AI. You will change your mind.

[Kevin O'Connor @kjpoconnor]

#appletv 3fa identification!? Absurd. TV, phone, and computer.

[Kevin O'Connor @kjpoconnor]

I hope @VivekGRamaswamy wins, but please STOP blowing up my text messages looking for money! I get 5/day from you.

[Kevin O'Connor @kjpoconnor]

@WinRed How do I get off your spam list? I mark all of your sms messages as spam, but you just use a different number. You're as bad as @actblue

[Kevin O'Connor @kjpoconnor]

Over 13 years ago, our last company, FindTheBest/Graphiq, received a demand letter from Lumen View alleging we had supposedly violated their patents. I decided to fight the patent troll, which led to a long and painful journey. I helped lobby Congress to pass patent reform laws. Other lobbyists were people whose companies and lives were destroyed by other patent trolls. We had bipartisan support, but the Bill was killed by Harry Reid, the Senate Majority Leader. Harry was owned by the trial lawyers who killed the deal. We continued our fight against the troll and were the first company to actually win an award against it (SCOTUS's Alice case set the precedent). Our company couldn't afford the legal battle, so I agreed to fund it. I spent $300k to win $100k. It was the best $200k I ever spent. https://t.co/CCwEi9lXfy

Along the way, I met Austin Meyer, who was even more incensed than I was. He funded a documentary in which he interviews people, including me, and explains how the scam works. The movie is called "The Patent Scam." https://t.co/g3wqCqy3DA

The scam works like this. A shell company spends $1 to send you a letter saying you violated their patent and they want $25k, or they will sue you. It only costs them $400 to sue you, but it costs you $3 million in legal fees to fight it. Even though most patents are invalid, you can't spend $3 million to invalidate them, so you pay up. I call it asymmetrical lawfare. Over the years, I've often contemplated how to level the playing field. I think I found a way.

Thanks to Claude Code, I was able to use AI to build a free tool that helps companies fight patent trolls: https://t.co/o2fzUYf6j5. AI analyzes the patent to find prior art and evaluate its "obviousness." You can even download a sample response to demand letters. In addition, the tool creates claims with diagrams that become prior art for those seeking to create derivative patents. Sort of a scorched earth approach. AI levels the playing field.

If you have received a demand letter in the past, please send it to me. Or, try adding a patent yourself to see how the tool works. Let me know what you think.

[Rogo @RogoAI]

Today, we’re announcing our $160 million Series D funding, led by @kleinerperkins, with participation from @sequoia, @ThriveCapital, @jpmorgan, and @khoslaventures.

35,000+ financial professionals. 250+ institutions. One platform, purpose-built for finance.

Thank you to our clients, our team, and our investors for their support as we continue building to raise the ceiling for what finance can accomplish.

[Max Spero @max_spero_]

Over the last year, I've watched a rise in AI content on basically every internet platform. Seeing a viral AI-generated post used to be a rare find. Now it's a daily occurrence.

Four months ago, we launched the @pangram bot to help people check long posts and articles for AI slop without leaving the platform. And it blew up. We went from a niche tool used by academics to a core piece of cognitive security infrastructure.

Today, we're taking it one step further. We're launching a Chrome extension that proactively scans all social content as you scroll, flagging AI content in real time so you can save your attention for what really matters: content authored by humans.

At launch, the Pangram Chrome extension will proactively scan posts on X, LinkedIn, Reddit, Substack, and Medium. And we'll give you a feed health summary, so you can see exactly which accounts are putting AI slop on your feed.

I'm so excited to share this with you all, and I hope you find it as useful as I do.

[Kevin O'Connor @kjpoconnor]

@RogoAI Awesome add, looking forward to using it!

[Rogo @RogoAI]

We are introducing Felix.

Felix is a purpose-built agent for high finance, designed for long-running, complex workflows and capable of producing decks, models, and documents end-to-end.

Felix executes so you can focus where it matters.

[Ivan Bercovich @neversupervised]

I want to share a new dataset of 331 reward-hackable environments. These are real environments used in Terminal Bench and adjacent benchmarks. I first got interested in this because, as a reviewer of Terminal Bench, I noticed a lot of our tasks were hackable. I also noticed that many contributors to the benchmark do so because it provides credibility when selling environments to labs. Hence, TBench tasks are, in my opinion, held to a higher quality standard than those being used today for RL. No one is spending hours manually reviewing the $1B in tasks being purchased by major labs. As far as I understand, while everyone knows environments are hackable, nobody has released hundreds of "realistic" environments. (link in comment)

[Kevin O'Connor @kjpoconnor]

@pmarca Odd, our partner at @scopvc, Ivan Bercovich, was banned by X for posting a Claude counterargument to @pmarca's Claude-generated argument. @elonmusk I thought your goal was to get rid of censorship!? I can't even imagine why it was censored in the first place. Bad AI algos?

[Kevin O'Connor @kjpoconnor]

@pmarca Odd, our partner at @scopvc, Ivan Bercovich, was banned by X for posting a Claude counterargument to @pmarca's Claude-generated argument. @elonmusk I thought your goal was to get rid of censorship!? I can't even imagine why it was censored in the first place. Bad AI algos?

[Kevin O'Connor @kjpoconnor]

@VerizonSupport Only took 1.5 hours to resolve. I spent 2.5 hours activating a phone!? You must activate 100k phones/day. Your CTO should contact me - you could save billions through simple automation.

[Kevin O'Connor @kjpoconnor]

Curious @Verizon, does your CTO @YagoTS
or anybody in engineering actually use your product? Your website doesn't work, your "AI" customer service is terrible, and nobody answers your phone. I tried to activate a new phone, but it failed after 3 attempts! I'm baffled how you don't know your system is completely broken. Please try harder!

[Ivan Bercovich @neversupervised]

https://t.co/buOPNmWeBe

[TheDataProject.ai @TheDataProject0]

🚨 NEW: OpenLobby — Follow the Money in Washington

We analyzed 650,000+ federal lobbying filings from 2018-2025.

$15.2 BILLION in lobbying income. Every client. Every lobbyist. Every issue.

What we found:
• 5,000 former govt officials now lobby Congress
• McKesson spent $1.45M lobbying → got $11.8B in contracts (8,187x ROI)
• Big Pharma spent $452M lobbying on drug pricing
• Big Tech spent $150M+ fighting AI regulation
• Foreign govts from 50+ countries lobby the US
• 2025 lobbying hit $2.7B — an all-time record

Built tools nobody else has:
🔢 Lobbying ROI Calculator
⚡ Influence Score Rankings
🔥 Real-time Surge Tracker
🔗 Cross-dataset intelligence (connected to our Medicare, Medicaid, and federal spending data)

All from public Senate LDA filings. No paywall. No login. No ads.

https://t.co/k2oWZ5X4jj

#lobbying #OpenLobby #government #DOGE #transparency #data

[TheDataProject.ai @TheDataProject0]

I spent a week digging through 227 million Medicaid billing records. $1.09 trillion in taxpayer money from 2018-2024. All public data from HHS that nobody was really looking at.

Built a site to make it all searchable. Here's what jumped out:

🚩 CARES INC (New York) — went from billing $1.6M in 2018 to $112.6M by 2023. That's a 6,886% increase. Seven separate fraud indicators triggered on them.

🚩 SRH CHN Lead Health Home LLC (New York) — a brand new entity that appeared out of nowhere and immediately started billing $239 million. Four flags triggered.

🚩 City of Chicago — $23M to $240M, a 942% spike. Three flags.

🚩 Massachusetts DDS entities — billing 37-51x the national median for residential habilitation services. Not a typo. Thirty-seven to fifty-one times.

Brooklyn has 64 flagged providers billing a combined $13.7 billion. All five NYC boroughs together — 111 flagged providers. New York is on another planet.

Arizona has a suspicious cluster of brand-new providers that popped up right after the pandemic and started billing aggressively. 46 new entrants flagged.

COVID was a gold mine for some — procedure code U0003 (COVID testing) generated $3.9 billion in Medicaid payments alone. LabCorp billed $174 million just in COVID tests.

The scariest part might be the small providers nobody's watching. One billed $652K with a fraud risk score of 0.898 out of 1.0. These aren't big hospital systems pushing boundaries — these are tiny operations with billing patterns that look nothing like legitimate practice.

Vermont has the highest per-capita fraud flag rate in the country. The most expensive single procedure costs $92,158 per claim.

I ran 13 statistical tests and a machine learning model against every provider in the dataset. Cross-referenced the OIG exclusion list. Built specialty-level peer benchmarking.

1,860 providers flagged. $226 billion in flagged spending.

Every provider is searchable. Every flag is documented with full methodology. All open, all free.

If you're a journalist, researcher, or just a taxpayer who wants to know where the money goes — dig in. And if you find something, let me know.

https://t.co/TJJfRh5vsB

[Kevin O'Connor @kjpoconnor]

@lukethomas14 You can look up the owner or admin names for those "companies" at: https://t.co/5yJybXeeox

Notice any patterns?

[Kevin O'Connor @kjpoconnor]

@aakashgupta @pangram

[Kevin O'Connor @kjpoconnor]

@pangram

[Kevin O'Connor @kjpoconnor]

Here is a brief guide on how to use the Pangram Labs bot on X (formerly Twitter) to detect AI-generated content.

How to Spot AI on X with @pangram

In an era where "dead internet theory" feels less like a conspiracy and more like a daily reality, distinguishing between a human thought and a ChatGPT output is becoming a necessary skill.

Enter Pangram Labs, a detection company that has released a simple, public-facing bot on X to help users verify the authenticity of posts in real-time.

How It Works

Using the tool is straightforward. When you see a post (tweet) that feels suspicious—perhaps the tone is too robotic, the structure too formulaic, or the opinion too generic—you can summon the bot directly in the replies.

Reply to the Post: Click the reply icon on the tweet you want to analyze.

Tag the Bot: Type @pangram (or sometimes simply !pangramlabs) and hit reply.
Wait for Analysis: Within a few seconds to a minute, the bot will reply to you with a score.
Interpreting the Score

The bot doesn't just give a "Yes/No" answer; it provides a probability percentage.

High AI Score (e.g., 98% AI): The text strongly matches patterns found in Large Language Models (LLMs) like GPT-4 or Claude. These posts often lack idiosyncratic human errors, use predictable sentence structures, and have a "flattened" emotional tone.

High Human Score (e.g., 99% Human): The text contains specific nuances, irregularities, or "burstiness" (variations in sentence length and complexity) that are characteristic of human writing and difficult for current AI models to replicate perfectly.
Why It Matters

As engagement farming becomes automated, reply sections are increasingly filled with AI bots designed to farm likes or push specific narratives. Tools like Pangram Labs provide a layer of transparency, allowing users to "fact-check" reality and filter out synthetic noise from genuine human discourse.

Note: Like all detection tools, it is probabilistic, not magic. It works best on longer posts (like threads) where there is enough text to analyze, rather than short, generic one-liners.