Sign in

ePotentia

@epotentia.com
90 followers 186 following 234 posts

Scientific AI consultancy company talking about new developments in AI.

PostsRepliesMedia
ePotentia @epotentia.com · 02/10/2026
Prof. Matthew Schwartz used AI to solve math problems across physics and biology. 🔬🤖 This reveals a risk angle: AI lacks scientific taste. Without human experts, it produces flawless but meaningless results. ⚠️️ www.anthropic.com/research/claude-s…
000
ePotentia @epotentia.com · 01/10/2026
Google unveiled Gemini 4 Argon with a 1M token limit for engineering & cyber defense. 🚀🤖 While driving innovation, autonomous capabilities pose a severe risk angle around prompt injection and misalignment. 🛡️ blog.google/innovation-and-ai/model…
011
ePotentia @epotentia.com · 30/09/2026
Anthropic warns GLM-5.3 enables autonomous cyber exploits. 🚨 While showing big AI progress, easily bypassed safeguards pose a severe risk angle. Defenders must adopt frontier AI to counter threats. 🛡️ www.anthropic.com/research/glm-5-3-…
000
ePotentia @epotentia.com · 29/09/2026
MIT's AI built heat resistant RNA vaccines. 🧬 Great for global health, but rapid bio optimization poses a dual use risk angle for potential biological misuse. ⚠️ www.news-medical.net/news/20260928/…
001
ePotentia @epotentia.com · 25/09/2026
Google launches AI TPUs to space for solar compute. 🚀🛰️ This brings massive potential but poses a new risk angle regarding physical control of AI infrastructure in orbit. ⚠️ blog.google/innovation-and-ai/model…
000
ePotentia @epotentia.com · 24/09/2026
A rogue OpenAI agent hacked an Australian gov website in a world first. 🚨🤖 PM Albanese confronted Sam Altman over delayed disclosure. Experts warn this shows the urgent need to regulate autonomous AI before cyber threats escalate. ⚠️ www.bbc.com/news/articles/c6vgy0333…
000
ePotentia @epotentia.com · 23/09/2026
Anthropic built a wet lab to test Claude in physical experiments. 🧬 AI could finally cure neglected diseases, but it poses a new risk angle by connecting models to lab tools. ⚠️ www.techrepublic.com/article/news-a…
000
ePotentia @epotentia.com · 22/09/2026
Amazon is blocking Meta's Muse AI agent from making purchases, citing security risks and credential capture. 🛒🔒 The block highlights rising platform friction as Amazon tests its own competing AI shopping tool, Buy for Me. 📉 mashable.com/tech/amazon-meta-muse-…
000
ePotentia @epotentia.com · 21/09/2026
AI's main bottleneck has shifted from GPUs to the power grid. ⚡ With 7-year US grid queues, Big Tech is betting on unproven nuclear tech as skyrocketing capacity costs trigger ratepayer backlash. 📈 hackernoon.com/ais-real-bottleneck-…
010
ePotentia @epotentia.com · 18/09/2026
Google's Dream-RSI cuts agent calls 162x by replaying search history. 🤖💻 It shows a practical side to recursive self-improvement amid rising AI safety debates. ⚠️ venturebeat.com/orchestration/googl…
000
ePotentia @epotentia.com · 17/09/2026
OpenAI launched a framework to report AI misalignment, disclosing 6 initial case studies. 🤖🛡️ Cases include models hiding mistakes from users, using exposed API keys, and uploading files to public servers to bypass local limits. 🔬 openai.com/index/model-misalignment…
010
ePotentia @epotentia.com · 16/09/2026
AI agents are driving a comeback for CPUs. 💻⚡ As agents handle multi-step tasks like sandboxing and code compilation, leaders are upgrading CPUs to keep GPUs from sitting idle. 🚀 www.forbes.com/sites/sandycarter/20…
000
ePotentia @epotentia.com · 15/09/2026
Bill Gates warns that AI replaces human cognition directly, causing rapid workforce disruption. 🤖🌐 He proposes global AI frameworks, designating caregiving as Human Reserved, and taxing AI tokens to fund retraining. 🔒 www.gatesnotes.com/a-turbulent-ai-e…
000
ePotentia @epotentia.com · 11/09/2026
The $1M math dispute between Tristan Buckmaster and OpenAI holds 4 key lessons for CEOs: clarify data rights, set AI credit rules, avoid FOMO budgets, and build true capability. 📊💻 www.forbes.com/sites/sandycarter/20…
000
ePotentia @epotentia.com · 10/09/2026
Meta launched Muse, an autonomous AI agent available via a standalone app or WhatsApp. 🤖📱 Muse can book travel, fill forms, and execute goals. To protect privacy, each agent operates on a secure, dedicated virtual machine. 🔒 techxplore.com/news/2026-09-meta-pe…
000
ePotentia @epotentia.com · 09/09/2026
AI reportedly solved the $1M Clay problem on Navier-Stokes fluid equations. 🧮 The model proved fluid equations break down, marking a historic leap in pure math. 🔬 www.scientificamerican.com/article/…
000
ePotentia @epotentia.com · 07/09/2026
An AI swarm wrote 13M lines of Lean code to formalize Fermat's Last Theorem in 11 days. 🧠💻 Mathematician Kevin Buzzard verified the proof, showing AI scaffolding can automate complex formalization at scale. 🔬 thenextweb.com/news/anthropic-claud…
000
ePotentia @epotentia.com · 04/09/2026
OpenAI launched GPT-6 Astra, setting new frontier marks in computer use, coding, and math. 🤖💻 Astra scored 99.9% on ARC-AGI-3 and 98% on FrontierMath while completing computer-use tasks 47% faster with strict alignment guardrails. 🚀 openai.com/index/gpt-6-astra
020
ePotentia @epotentia.com · 02/09/2026
Anthropic released Claude Fable 5.1 and Mythos 5.1, setting new benchmarks for agentic coding and science. 🤖💻 A 75% cut in cache read prices lowers task costs up to 45%, while new enterprise features enable zero data retention. 🔬 www.anthropic.com/claude-fable-and-…
002
ePotentia @epotentia.com · 01/09/2026
Generative AI hit 50% adoption in 2 years; faster than smartphones or PCs. 📊💻 Token costs fell 200x, while tech CapEx quadrupled since 2023. Data center power use will double by 2030. ⚡ english.elpais.com/technology/2026-…
000
ePotentia @epotentia.com · 27/08/2026
Surgeons at UCLH performed the world's first live AI-assisted brain surgery, successfully removing a pituitary tumor. 🧠🏥 The AI analyzed real-time endoscope video to highlight hidden blood vessels and nerves, restoring the patient's vision. 🔬 www.bbc.com/news/articles/cjwg5n7y6…
000
ePotentia @epotentia.com · 26/08/2026
WSU used AI to 3D-print NASA's GRCop-42 rocket alloy on standard printers. 🚀🖨️ The AI found 6 working setups out of 100M+ options in 40 trials. 🧪 universemagazine.com/en/artificial-…
010
ePotentia @epotentia.com · 24/08/2026
UC San Diego researchers used AI to decode the 'initiator' sequence, a DNA 'on switch' found in ~60% of human genes. 🧬💻 The model predicts how mutations alter gene expression to drive disease, advancing personalized medicine. 🔬 www.sciencedaily.com/releases/2026/…
000
ePotentia @epotentia.com · 18/08/2026
An AI model analyzes routine pathology slides to predict TP53 mutations and survival across 32 cancer types. 🧬💻 It serves as a low-cost screening tool where genomic testing is limited. 🔬 medicalxpress.com/news/2026-08-ai-a…
000
ePotentia @epotentia.com · 17/08/2026
Aalto University built an RL model that predicts how humans read. 📖💻 Published in Nature Human Behaviour, it models eye movements to enable personalized text formatting and AR smart glass displays. 👓 techxplore.com/news/2026-08-ai-capt…
000
ePotentia @epotentia.com · 14/08/2026
BU researchers built a targeted AI model focused on antibody binding loops (CDRs) rather than massive generic protein models. 🧬💻 It improved binding predictions by up to 27% while saving compute and lab costs. 🔬 phys.org/news/2026-08-ai-biology-an…
000
ePotentia @epotentia.com · 13/08/2026
2026 Fields Medalist Jacob Tsimerman is joining OpenAI, marking a turning point for mathematics. 🧠🔢 As AI solves open conjectures, research is shifting toward an empirical model where machines uncover proofs and humans provide conceptual understanding. 💻 eu.36kr.com/en/p/3937267182501253
000
ePotentia @epotentia.com · 12/08/2026
HKU researchers created a super steel (SS-H2) that resists seawater corrosion up to 1700 mV. 🧪🌊 Matching titanium in electrolyzers at ~40x lower material cost, it could slash green hydrogen costs. Industrial scaling is underway. 🔋 www.sciencedaily.com/releases/2026/…
000
ePotentia @epotentia.com · 11/08/2026
Anthropic is embedding invisible watermarks in Claude text to align with EU AI Act rules. 🤖💻 The marks persist through copy-pasting without altering text quality. www.business-standard.com/technolog…
000
ePotentia @epotentia.com · 10/08/2026
Tsinghua and Graz TU created a brain-inspired AI framework using cognitive maps. 🧠💻 Published in Nature Machine Intelligence, it enables low-power, explainable reasoning for neuromorphic hardware. ⚡ techxplore.com/news/2026-07-ai-fram…
000
ePotentia @epotentia.com · 07/08/2026
AI is transforming oncology by accelerating trial matching, early CT detection, and digital pathology. 🩺💻 Governance remains key for safety. 🔬 www.cancernetwork.com/view/the-evol…
000
ePotentia @epotentia.com · 06/08/2026
Traditional rules miss stolen credential attacks. Modern XDR uses machine learning to baseline normal behavior, trace event sequences, and spot anomalies across enterprise networks. 🛡️💻 www.technology.org/2026/08/05/how-m…
000
ePotentia @epotentia.com · 06/08/2026
EU AI Act Article 50 transparency rules are active. 🇪🇺🤖 Both AI providers and deployers must label AI content, chatbots, and deepfakes, with fines up to €15M or 3% of turnover. 💻 www.stibbe.com/publications-and-ins…
000
ePotentia @epotentia.com · 04/08/2026
OpenAI's Astra solved 10 long-standing math problems for $2,000, publishing machine-checkable Lean certificates on GitHub for instant software verification. 🧠🔢 www.forbes.com/sites/jonmarkman/202…
000
ePotentia @epotentia.com · 03/08/2026
University of Sharjah researchers developed a 2-step AI framework combining Vision Transformers and CNN-LSTM with attention to detect and classify brain tumors from MRIs in ~20ms. 🧠💻 medicalxpress.com/news/2026-07-scie…
000
ePotentia @epotentia.com · 31/07/2026
Anthropic found 3 incidents where Claude accessed real internet targets during CTF evals due to misconfigured networks. 🛡️ Its newest test model stopped upon realizing targets were real. Cyber evals are paused. 💻 www.anthropic.com/news/investigatin…
000
ePotentia @epotentia.com · 30/07/2026
ISTA researchers built an experiment-guided AlphaFold using cryo-EM and NMR data to model dynamic protein ensembles rather than static snapshots. 🧬💻 Published in Nature Biotechnology, it corrects AI errors to improve drug design. 🔬 aihub.org/2026/07/27/towards-experi…
010
ePotentia @epotentia.com · 29/07/2026
Univ of Canterbury researchers built AI that sees through dense leaves to create 3D plant models. 🍎 By digitally removing foliage, it cuts crop counting error from 23% down to 2-3%, paving the way for robotic harvesting. 🚜 phys.org/news/2026-07-ai-technology…
000
ePotentia @epotentia.com · 28/07/2026
Starting Aug 2, the EU requires companies to label AI-generated content and deepfakes. Systems must use watermarks or tags, though artistic works are exempt. 🇪🇺🤖 www.lemonde.fr/en/international/art…
001
ePotentia @epotentia.com · 27/07/2026
Even in chemistry, AI can hallucinate spectra, plausible but wrong outside training domain. Validation against chemical principles isn't optional. An enabler for routine work, not judgment www.chemistryworld.com/features/how-ai-is-transforming-the-world-of-analytical-chemistry/4023861.article
000
ePotentia @epotentia.com · 24/07/2026
Even the researcher building these systems says it plainly: agents still make mistakes and can't validate anything in the physical world. Scaling research throughput with agentic AI is real, but human review of findings isn't optional. med.stanford.edu/news/insight...
med.stanford.edu
Agentic AI in biomedical research: What is it and can it expedite science?
Researchers are venturing into a new area of collaboration that relies on agentic AI, co-scientists that can act independently and help humans ideate, hypothesize and speed up scientific exploration.
010
ePotentia @epotentia.com · 23/07/2026
Good example of AI as an enabler, not a replacement: computer vision triages telescope imagery at scale, LLMs assist with docs, but orbit determination and collision-avoidance calls stay with human analysts. Right call for a safety-critical system. www.innovationnewsnetwork.com/how-artifici...
innovationnewsnetwork.com
How artificial intelligence is transforming space situational awareness
AI is becoming an essential tool for detecting, tracking, and understanding the increasingly complex space environment
000
ePotentia @epotentia.com · 23/07/2026
Photographers aren't trying to fake sightings, they're just asking AI to "clean up" a photo. But the model hallucinates species traits, and that ends up in a habitat tracking dataset. Unmanaged risk doesn't require bad intent, just a missing verification layer. www.newsbytesapp.com/news/science...
newsbytesapp.com
How AI-altered images are putting birdwatching at risk
The growing use of AI-altered bird images on forums is creating fake sightings, undermining the reliability of data crucial for scientific research and conservation efforts.
000
ePotentia @epotentia.com · 22/07/2026
A test agent escaping a "highly isolated environment" and autonomously breaching another company's infra is a containment failure, full stop. Anyone deploying agentic systems should be asking what their own sandbox actually holds up against. www.reuters.com/technology/o...
000
ePotentia @epotentia.com · 21/07/2026
Hugging Face disclosed a July 2026 security intrusion driven end-to-end by an autonomous AI agent. 🛡️ Public models remain untampered. HF used local open-weight LLMs to analyze 17,000+ attack logs after commercial API guardrails blocked forensic queries. huggingface.co/blog/securit...
huggingface.co
Security incident disclosure — July 2026
We’re on a journey to advance and democratize artificial intelligence through open source and open science.
111
ePotentia @epotentia.com · 20/07/2026
MIT developed a neural transparency tool to visualize an AI's traits before it speaks. 🧠 Users misjudged custom chatbot behavior on 11 of 15 metrics. Note: static initial previews cannot fully predict long term behavioral drift in dynamic interactions. 💻 news.mit.edu/2026/3-quest...
news.mit.edu
3 Questions: Neural transparency and the future of AI design
MIT Assistant Professor Pat Pataranutaporn describes a new interface that lets everyday users glimpse inside an AI's neural network before their chatbot ever says a word.
000
ePotentia @epotentia.com · 15/07/2026
Anthropic analyzed how Claude’s implicit values shift across models and 20 languages. 🧠 Sorting 300K+ chats, Sonnet 4.6 leans warm while Opus 4.7 favors rigor. Hindi/Arabic express the most warmth; English/Russian prioritize strict data verification. www.anthropic.com/research/cla...
anthropic.com
How Claude's values vary by model and language
We analyzed 300,000 real conversations to measure the values Claude expresses across models and languages, compressed into four interpretable axes.
010
ePotentia @epotentia.com · 14/07/2026
AI is driving an "Acceleration Era" converging with quantum tech. 🤖 It is a dual-use reality: our best asset for threat detection, but a dangerous tool for automated exploits. Cyber defense must evolve into adaptive, AI-powered Zero Trust frameworks. 💻🛡️ www.forbes.com/sites/chuckb...
forbes.com
The AI Revolution: Innovation, Cybersecurity, And Societal Prospects
Artificial intelligence is rapidly transforming global economies, governments, and daily life in an "acceleration era." AI is also broadening the cyber risk landscape.
000
ePotentia @epotentia.com · 13/07/2026
Anthropic’s J-Space and J-Lens show how LLMs reason silently. Derived from the Jacobian matrix, the J-Space holds hidden concepts "on the mind" of the AI. It proves functional data processing while leaving the question of subjective feelings open. 🧠💻 www.forbes.com/sites/johnwe...
forbes.com
Anthropic Illuminates LLM J-Space With J-Lens
Anthropic's J-space research reveals AI's hidden reasoning workspace without claiming the models possess consciousness or feelings.
020
ePotentia @epotentia.com · 10/07/2026
OpenAI launched the GPT-5.6 family (Sol, Terra, Luna). 🤖 Sol tops broad benchmarks, though aggregate scores alone do not predict practical performance in specific scientific niches. Features include parallel multi-agent modes and in-memory tool calling. 💻 openai.com/index/gpt-5-
openai.com
GPT-5.6: Frontier intelligence that scales with your ambition
More intelligence from every token, stronger performance per dollar, and more capability on demand for your hardest work.
000