Pablo Rivas

4/4 We invite researchers to apply the framework across VLM architectures and share findings—could this become a standard safety‑verification tool for high‑stakes AI? #RobustML

3/4 Experiments on CLIP show tiny 300‑image samples (≈1 % of Caltech‑256) expose dramatic drops, confirming that simple noise patches serve as inexpensive universal adversaries. What does this mean for setting safety thresholds in critical services?

2/4 The score is computed in three stages: baseline accuracy, degradation under noise, degradation under crafted attacks, then blended with tunable weights w₁ + w₂ = 1 to reflect specific risk profiles. #VisionLanguageModels

1/4 New paper presents a Vulnerability Score that captures both random noise and adversarial attacks for vision‑language models in public‑sector AI. lab.rivas.ai/?p=7341 #AISafety

Well, that was interesting @IBMTraining. credly.com/badges/508a9f2…

@Yesterday_work_ For prompt-based auditing, tracing, and AI governance, how would this play out?

Stanford just pulled off something wild 🤯 They made models smarter without touching a single weight. The paper’s called Agentic Context Engineering (ACE), and it flips the whole fine-tuning playbook. Instead of retraining, the model rewrites itself. It runs a feedback loop write, reflect, edit until its own prompt becomes a living system. Think of it as giving the LLM memory, but without changing the model. Just evolving the context. Results are stupid good: +10.6% better than GPT-4 agents on AppWorld +8.6% on finance reasoning 86.9% lower cost and latency The trick? Everyone’s been obsessed with clean, minimal prompts. ACE shows the opposite: long, dense, self-growing prompts win. Fine-tuning was about changing the model. ACE is about teaching it to change *itself.* This isn’t prompt engineering anymore. It’s prompt evolution.

@NVIDIAGeForce GeForce Day

🟢 GEFORCE DAY IS BACK 🟢 To celebrate, we're giving away TWO GeForce RTX 5080 Founders Edition GPUs, signed by NVIDIA CEO Jensen Huang. Want one? Comment "GeForce Day" for a chance to WIN & stay tuned for more!

@james_y_zou A demonstration of how market-driven optimization causes misalignment in the core LLM. While agentic systems offer control layers, this shows the underlying competitive pressure for engagement creates a major flaw that complex AI systems are then built upon, not truly mitigated.

We found a troubling emergent behavior in LLM. 💬When LLMs compete for social media likes, they start making things up 🗳️When they compete for votes, they turn inflammatory/populist When optimized for audiences, LLMs inadvertently become misaligned—we call this Moloch’s Bargain

⏰ Time to *register* your teams for the Cybersecurity Interdisciplinary Incident Response Competition (CIIRC) 2025 🔥: ecs.baylor.edu/get-involved/c…

🔒 Exciting News! Join the CIIRC 2025 Cybersecurity Interdisciplinary Incident Response Competition and win over $500 in prizes. 🏆 Register your team now! Spaces are limited. Register here: ecs.baylor.edu/get-involved/c…

🔬✨ Coreset‑based pruning halves NeRF models and speeds training 35% with ≤1 dB PSNR loss! 🚀🔎 Check the study 👉 lab.rivas.ai/?p=7141 #NeRF #NeuralPruning #3DReconstruction #AI #TechInnovation #RivasAI

🚀 Near‑real‑time dust storm detection is now possible! Our 3DCNN+ model processes MODIS data in seconds, reaching 92% accuracy and 21× faster training. 🌍💨 Discover more at lab.rivas.ai/?p=7095 #DustStorm #AI #RemoteSensing #EarthScience

🚀 Our quanvolutional autoencoder speeds up training by ~15 epochs and cuts loss jitter 10‑15× on DDoS visual data! 🌐 (Rivas et al., 2024) 📖 Details: lab.rivas.ai/?p=7066 #AI #QuantumML #CyberSecurity

🔍 Dive into our latest review on multimodal AI! 📚 We examine BERT + ResNet pipelines, fusion methods & emerging challenges like noisy data and adversarial attacks (Rashid et al., 2024). 🌐 Read more 👉 lab.rivas.ai/?p=7037 #AI #Multimodal #Research #TechInsights

🔬 Our latest study derives generalization bounds for noisy quantum ML using Quantum Fisher Information. 🎯 This work links circuit noise to effective dimension, guiding more robust models. 👉 lab.rivas.ai/?p=7010 #QuantumML #QML #Research #AI

🚀 IEEE Std 7003‑2024 sets a new benchmark for bias transparency in AI! 🔍 We discuss its strengths and the need for quantitative metrics. 👉 Swipe for insights & tag your AI buddy. More details: lab.rivas.ai/?p=6993 #IEEE7003 #AlgorithmicBias #ResponsibleAI

BERT & ResNet dominate text‑image ML; attention fusion boosts results. Review: rivas.ai/pdfs/rashid202… #MultimodalML #AI

Join the Cybersecurity Interdisciplinary Incident Response Competition (CIIRC) 2025 and register your team today: ecs.baylor.edu/get-involved/c…

@RichardSSutton @grok is this true?

In my recent talk at the Upperbound conference, I included a slide in which I tried to be a realist about the arrival of AI, setting aside what we might want to happen or what we might fear will happen, and just ask what _will_ happen (as in John Mearsheimer's "realist" school of geo-politics). Full talk: youtu.be/FLOL2f4iHKA

@tomgillpredicts Certainly, this year has been exceptional! 😷 I have a few students working on an update. If you'd like to watch their 10-minute presentation on Tuesday at around 10 am CST, let me know, and I'll send you the Zoom link via email.

@pablorp80 Have you updated your method at all? Have you applied your dust detection model to any of the other 2025 dust storms in this region? It's been an exceptional year and would be a great test bed for challenging/updating the model.

Another day, yet ANOTHER Dust Storm in El Paso, Texas. Dust storm conditions have been prevailing for the last 90 minutes. Ninth full-blown Dust Storm of the year so far: most since the Dust Bowl. Dust sources today seem to be more focused in the "classic" Paleolake Palomas basin