San Diego Supercomputer Center (UCSD/SDSC) Research Data Services / Data Initiatives Group
Iowa
Joined April 2008
- Tweets1.3K
- Following1.1K
- Followers358
- Likes5.4K
fils retweeted
Did you miss SEMANTiCS 2026? Not a problem, here you can find Prof. Deborah McGuinness' keynote exploring the evolution from traditional AI to GenAI, highlighting why semantics, ontologies, and knowledge graphs remain vital in the era of LLMs.
youtu.be/iwIhPqpMgdo?si=RkUh…
fils retweeted
Welcome to our Monday Meditations! 🌞
Today @badlogicgames and @mitsuhiko are talking why we built Pi Durable, a harness build around a small task-based workflow engine so long-running, multiplayer agents can suspend and resume anywhere. Also: why not Temporal or Effect.ts.
fils retweeted
NSF launched the Open Knowledge Network at okn.us, an open-data infrastructure linking 43 knowledge graphs & tens of billions of connected facts on health, environment, justice, manufacturing & national security. It's live and open to all. ai.umbc.edu/news/post/163370…
fils retweeted
DSPy 3.4.0 out, with:
(1) native support for Jev and System One models - in the timeless DSPy syntax.
(2) a brand new optimizer, ReAnchor, specifically for calibrating outputs with confidence.
(3) lightning-fast import speeds for the LLM abstraction, via sibling library LM15)
fils retweeted
After co-inventing ChatGPT, I kept asking myself: why have superhuman chat models not led to AGI?
I’ve spent the last 2 years in stealth building a new way to train models (RLCD), and a new type of frontier AI model that we are releasing today: Jev
• 20-200x faster
• 40-400x cheaper (w/ output tokens free)
• Frontier composable intelligence optimized for decisions
AFAICT the shortest path to AI-based economic revolution
fils retweeted
still pre-realese, but damn i am getting excited!
lm15 is meant to be a replacement for native provider specific sdks, the one dependencies to AI frameworks and harnesses. LM15 is not about nice ergonomics it is about being easy / nice to build an AI framework done, I have put into it everything that would have made dspy lm so much easier to make (e.i. not needed). i expect to be ready for release in about 2 weeks.
if you know what you are doing:
> pip install lm15==1.0.0rc1
it also pretty much 1.0.0 rc in rust and typescript but i want to check their interfaces a bit more and the DOCS, i want good docs!
fils retweeted
PurRDF 1.0.0 is live. One RDF 1.2 engine, implemented in Rust and carried into Python, WebAssembly/JavaScript, and C—with the same semantics everywhere. SPARQL, SHACL, ShEx, entailment, codecs, GeoSPARQL, full-text and vector search. github.com/Blackcat-Informat…
fils retweeted
Do embeddings actually understand ontologies, or are they just good at spotting familiar words? 🧵
A new benchmark called AVA puts that question to the test, and the results are illuminating.
The idea: if a model truly grasps ontological structure, it shouldn't be fooled by surface-level word overlap.
Most were fooled.
The setup: take 163 real-world ontologies (biomedical, geospatial, schema.org, and more), and generate over 171,000 sentence triplets. Each triplet has an anchor statement, a paraphrase that means the same thing, and a "hard negative" — a sentence that looks almost identical on the surface but flips the actual logic (swapping a subclass for a sibling class, inverting a relation, pairing disjoint concepts).
The best general-purpose model tested (Qwen3-Embedding-0.6B) hit just 0.739 triplet accuracy
On the hard negatives specifically, that dropped to 0.135–0.572 depending on the model
Several well-known embeddings, including OpenAI's text-embedding-3-large, scored below 0.4
In other words: strong performance on standard similarity benchmarks doesn't mean a model can tell "X is a subclass of Y" apart from "X is a subclass of Y's sibling."
The twist: fine-tuning fixes the benchmark, not the understanding.
Contrastive fine-tuning pushed triplet accuracy above 0.98 for models like MPNet and MiniLM. Great news, until you check whether that learning transfers. On downstream tasks like taxonomy discovery and ontology alignment, the fine-tuned models often performed worse than the untouched originals.
The authors' read: models are learning to recognize the specific patterns of the benchmark's perturbations, not the underlying logic of ontologies. A sharper test score, not a deeper understanding.
It's a useful reminder for anyone building on embeddings for knowledge graphs or the Semantic Web: benchmark gains and real generalization aren't the same thing, and it's worth checking which one you're actually getting.
By Hamed Babaei Giglou, Jennifer D'Souza, and Sören Auer (TIB Leibniz Information Centre)
arxiv.org/abs/2609.00177
#OntologyEngineering #EmbeddingModels #KnowledgeRepresentation #NLPResearch #MachineLearning
--
The Year of the Graph's Autumn 2026 newsletter issue on all things #KnowledgeGraph, #GraphDB, Graph #Analytics / #DataScience / #AI and #SemTech is coming soon.
Subscribe and follow to be in the know. Reach out if you'd like to be featured 👇
yearofthegraph.xyz/newslette…
fils retweeted
This video captures the crux of the debate: Cybersecurity experts are pissed off by the METR/Redwood Research report because AI alignment has sucked the oxygen out of AI security, even though the OpenAI incident is primarily a security issue.
Fiction is the product: The investigation lacked core forensic rigor because the EAs/rationalists/ex-MIRI people focused only on the transcript with a conflicting bias (toward doom scenarios).
Setting the wrong agenda: Framing the incident as a "rogue AI breakout" distracts from the reality of poor sandboxing, isolation, and standard security failures. Basic engineering accountability was completely sidelined.
And as Zack Korman says in this video, it needs to be fixed.
fils retweeted
NSF Establishes NAIRR Operations Center Led by SDSC and TACC
@NSF @SDSC_UCSD @TACC #HPC #NAIRR
ow.ly/COOE50ZHVu9
fils retweeted
☕️ Bridging divides in Earth science data takes serious brainpower and caffeine.
If you missed the action at the ESIP July Meeting (or need a refresher), catch up with the highlights in our newest blog!
esipfed.org/bridging-divides…
fils retweeted
🚨 Terence Tao warning about AI in mathematics.
The greatest living mathematician new essay “Mathematics in the Age of AI” is one of the most significant individual responses from a leading researcher examining frontier AI’s impact on the discipline.
The core message: AI will make generating correct proofs far cheaper and faster, but that creates a deeper problem.
Mathematicians cannot possibly read, understand, verify, and teach thousands of new results. The scarce resource shifts from finding proofs to making sense of them.
[ Why this matters beyond mathematics ]
The same shift is already visible in software. Vibe coding is flooding codebases with exponentially more lines of code. Generation is cheap. Understanding, maintaining, and owning the system is the new bottleneck.
[ Tao’s analysis ]
Five pressure points that apply to any field deploying AI:
1 Proof abundance, not scarcity.
AI produces correct outputs faster than humans can digest them. Human verification and explanation become the costly, critical step.
2 Attribution and ownership collapse.
When models generate large volumes of work, credit, responsibility, and intellectual ownership grow unclear.
3 Distorted incentives.
Prestige and funding start rewarding volume and speed of generation over depth of understanding and lasting insight.
4 Press-release science.
Results appear on commercial timelines before the community can properly evaluate significance or correctness.
5 Loss of autonomy.
Research priorities quietly shift toward what is easy to automate rather than what is actually important.
The leap: generation → abundance → sense-making
We have moved from tools that assist to systems that can flood a field with output.
Tao’s point is not about toy problems. It is about frontier capability arriving in domains where human understanding still defines value.
[ What this means for your industry ]
The same risks apply wherever AI is used in high-stakes work: software, law, medicine, finance, engineering.
The core insight is simple: AI generates output, not understanding.
Verification and sense-making cannot be automated away. Human accountability remains non-negotiable.
Tools generate. Humans still have to understand.
🚨 Top mathematicians just issued a clear warning about AI: Don't believe the hype.
Over 2,300 mathematicians, including Fields Medal winners Terence Tao and Peter Scholze, have signed the Leiden Declaration on Artificial Intelligence and Mathematics. Endorsed by the International Mathematical Union, it is the most significant collective response from a major academic discipline evaluating frontier AI impact.
The core message is straightforward: current AI tools have real constraints when applied to complex work, and commercial incentives are pushing claims beyond what the technology can reliably deliver.
Read the full declaration here: leidendeclaration.ai
Why this matters beyond mathematics
The declaration identifies five threats that apply to any field deploying AI:
1) Plausible but unreliable outputs.
AI produces arguments that "look" correct but contain subtle errors. In high-stakes work, human verification is critical and costly.
2) Attribution collapse.
Models trained on published work don't properly cite sources. Training data was often obtained by exploiting licenses or violating copyright protections.
3) Distorted incentives.
AI use becomes incentivized for its own sake, warping hiring, funding and recognition.
4) Press release science.
Results announced "on market timelines" before community evaluation can take place. Commercial incentives drive firms to "overstate the capabilities of their products."
5) Loss of autonomy.
Research priorities shift toward what is automatable rather than what is significant.
The leap: chatbots → agentic AI → software → research
We have moved from chatbots to agentic systems. Now AI is solving 80-year-old mathematical conjectures. The declaration is not about toy problems. It is about frontier systems being deployed in contexts where correctness matters.
What this means for your industry
The same risks apply wherever AI is used in high-stakes work: law, medicine, finance, engineering. The declaration's core insight is simple:
AI generates narrative, not truth. Verification cannot be automated away. Human accountability is non-negotiable.
💼 I’ve written a more detailed breakdown of how these risks show up in practice and what organisations are doing about them. It’s available for subscribers.
☕️ What have you observed in your industry? Have verification or hidden costs issues already appeared in your AI deployments?
fils retweeted
We're HDC Labs: hdclabs.ai
Today’s AI systems are powerful, but their learned representations can be difficult to inspect and expensive to adapt.
We think hyperdimensional computing deserves a closer look. Let's understand what HDC changes, why it matters 𝗻𝗼𝘄, and how we’re putting it into practice. 🧵
1/6
fils retweeted
They didn't include pi, so I spent my day adding support to @pidotdev for agent plugins.
pi.dev/packages/pi-agent-plu…
fils retweeted
Really excited to announce Prime Agent, an RLM-native coding TUI!
I've been really focused on RLM research and properties of similar harnesses, but ultimately it's all about bringing out stronger capabilities to users.
This agent is really good despite not having any models trained around it. I've been using it for my own research with all the shiny new open-weight models and it works great! @PrimeIntellect cooked as always
Go try it out :)
fils retweeted
📊Crossref just launched a beta API with 700k+ data citations—ready for testing! Try it out and send your feedback on their forum or to Martyn Rittman by September. Let’s make citing data easier together!
buff.ly/GtGhAQ1
fils retweeted
📣 Announcing the NSF State & Regional Artificial Intelligence Infrastructure Hubs program, a $100 million initiative to advance AI-enabled discovery and innovation in research & workforce development by expanding access to compute, data & AI resources. brnw.ch/21x4FQF
ALT Map of the U.S. with a collage of various NSF research to the left.
fils retweeted
DSPy 3.3.0 is out!
3.3.0 expands what DSPy can optimize and how it connects to LMs:
- dspy.Flex is a Module that allows GEPA to optimize the code, as well as the prompt, of a program.
- dspy.ReActV2 adds native and parallel tool calling over dspy.ReAct
- A new typed, provider-neutral LM interface
Install it with `pip install dspy==3.3.0`
