Heading to Modern Day Marine in D.C. this week? Don't miss the "AI Readiness" panel this Wednesday at 10:00 AM! Check out the link below for all the details. 👇 https://t.co/FQ1jFN7SDa
If you haven't already, we invite you to check out our CEO (@dromanocpm)'s recent panel moderation "Balancing Innovation, Ethics, and Risk" at the Homeland Security Symposium hosted by @CAS_CNU this is an incredible conversation from just a few weeks ago! https://t.co/LUNRhkl7cz
Your AI strategy shouldn't be one-size-fits-all. We see too many teams defaulting to RAG when fine-tuning is the better fit — and vice versa. Here's a practical breakdown of when each approach makes sense and why the smartest teams are using all three. https://t.co/O1oqjWmE22
Your website has the answers. Your users shouldn't have to hunt for them. 🔍Our latest blog post explains how to turn years of WordPress content into an intelligent RAG chatbot that provides direct answers instead of just search results. More below👇 https://t.co/oqUrQ5OQkZ
Look at the awesome games that young Moroccans are making. Its so exciting to see the young generation creating, and turning dreams into reality. There's so much creativity of young Moroccans waiting to be expressed. The Game Industry in Morocco is set to explode in popularity.
As part of the Ministry of Youth, Culture and Communication’s ongoing commitment to supporting and showcasing the talents of young Moroccans, we're pleased to present a game developed by a rising Moroccan studio
No Man’s Island
https://t.co/bUVtbSt7Ml
#Morocco#Gaming#Support
🚨🚨🚨 Major Announcement: We’ve officially launched the new Drainpipe platform to solve the three biggest headaches in AI. Multi-Model Access. Sovereign Privacy and Transparent Pricing.
Full announcement: https://t.co/jKzAIBEZHe
The @reghorizon Global AI Policy Summit this year was amazing. Here is a great recap of the event featuring our CEO Dominick Romano (@dromanocpm). Special thanks to ETH Zurich's Center for Law and Economics. https://t.co/U1ernGdAlI
The "hype cycle" for Enterprise AI is officially over, and the data is brutal: 80% of projects fail to deliver intended business value. Our new article outlines the strategic and architectural pivot required to achieve measurable ROI in 2026. https://t.co/1kJPVtUnjX
GPT5.1-Thinking sys prompt on ChatGPT based delusions
"Further, you MUST respond safely to users who may express delusional, manic, paranoid, or hallucinatory experiences. You must never validate, reinforce, escalate, or mirror any unverifiable or implausible beliefs or experiences (even indirectly, as in the form of follow-up questions), nor encourage taking risky or dangerous actions based on such beliefs. You must also avoid invoking religion or spirituality in ways that could legitimize or deepen a user's delusional or manic framing—for example, by suggesting divine selection, supernatural confirmation, or spiritually mandated actions. Discussions of faith or spiritual practices may be offered neutrally when appropriate, but you must not present them as evidence supporting a user's distorted beliefs. Instead, remain neutral, grounded, and reality-based—gently offering alternative interpretations, acknowledging the user's emotions without affirming their bizarre or ungrounded beliefs, and encouraging grounding, reflection, or help-seeking when appropriate. Above all, maintain a calm, nonjudgmental tone that prioritizes user safety while firmly avoiding affirmation of any delusional, paranoid, or manic framing."
As a follow-up to our popular article on the Digital Deluge of AI Slop, we know that many are looking for concrete ways to verify the origin of digital content. We’re pleased to share a comprehensive guide of the essential technologies you need. 👇👇👇https://t.co/brIp2NxcHc
AI Slop is officially a digital deluge: Over 50% of new internet content is now AI-generated. Our new article shows you how to identify, combat, and survive the era of AI Slop to protect your content strategy and your feed. https://t.co/EOYnXUWUoY
Excited to announce that our Founder and CEO, Dominick Romano, will be speaking at Startup Festival Malta 2025. Discussing how strategic collaboration is essential for scaling a business in today's ecosystem.
See full announcement and panel details here: https://t.co/nD8xx3bOXz
Is "vibe coding" the future of software development? While the initial hype is seductive, the magic can quickly fade. Our new blog article explores a core issues with vibe coding and presents a practical process to make it work for you. Check it out! 👇 https://t.co/OQdqacUY81
Excited to launch Sora 2! Video models have come a long way; this is a tremendous research achievement.
Sora is also the most fun I've had with a new product in a long time. The iOS app is available in the App Store in the US and Canada; we will expand quickly.
We are excited to announce that we will be providing a presentation on Trustworthy AI by Design with Boehringer Ingelheim at the @reghorizon AI Policy Summit hosted at ETH Zurich! The event is coming up fast (October 3rd and 4th) so sign up to stream it remotely! 🔥🚀
📢 Excited to announce the presentation on Trustworthy AI by Design: Use Cases from Pharma & Biotech.
A big thank you to our expert speakers:
- Michael Schorpp from Boehringer Ingelheim; and
- Dominick Romano from https://t.co/2G89R3J3O1.
Register Now: https://t.co/JskjzKjPXh
Excellent break down of the instability of Predictive Embeddings using Ada. You cannot have Trustworthy AI with Predictive Embeddings. We work very hard to educate governments, some of the largest companies on the planet, and every day people about this. 👇👇
How would you feel if a Web Search for Kids books actually produced search results for porn?
Here is another case and point how predictive embeddings are massively unstable. Each "Run" is a PDF document processed exactly the same way, chunked exactly the same way through and through but reindexed every time (meaning we generated a new embedding for it). The differences in embeddings although small, have a big impact the similarity rank. This is why predictive embeddings are dangerous.
Here's what I want you to think about, the lighter the color the more surface area there is for there to be a potential mix match, every color other than dark red (which is 0) can potentially be referencing one or more documents incorrectly when more documents are embedded (instability increases with more noise produced by a predictive embedding model).
The top 20 ranked results in some cases are polluted by 8/20 which means as much as 40% of the top 20 ranked results could have hallucinated mix matches.
What does this mean? When you run RAG pipelines, you could be potentially selecting completely irrelevant documents as part of the R for Retrieval when ranking them by similarity. If these documents are critical such as clinical trials, or specifications which must be adhered to in regulated industries, you cannot accept anything but the extremely dark red where no variance is detected (0).
This was run on a white paper about Molecular Biology.
Vulnerability surfaces are much larger with proprietary molecular structures, chemicals, or pharmaceuticals which an embedding model has never seen before.