#Truth&Inference
New preprint, and my first as last author!

We built realistic simulated genomes where the true history of every gene is known, so orthology tools can be tested against the truth

www.biorxiv.org/content/10.6...

(1/5)
OrthoSimulator: Empirically trained simulations for ground-truth benchmarking of orthology inference
Orthology inference is foundational for comparative genomics, enabling the reconstruction of phylogenetic relationships and transfer of biological knowledge between species. Methods often infer differ...
www.biorxiv.org
November 16, 2026 at 9:59 AM
OrthoSimulator: Empirically trained simulations for ground-truth benchmarking of orthology inference https://www.biorxiv.org/content/10.64898/2026.11.12.766176v1
November 16, 2026 at 10:48 AM
Talking head says what goes on in private is very different from what goes on in public, inference being that trump will grow a pair in private.

Truth is, of course, things are MUCH worse for us with anything trump does in private!
November 19, 2026 at 3:10 PM
OrthoSimulator: Empirically trained simulations for ground-truth benchmarking of orthology inference https://www.biorxiv.org/content/10.64898/2026.11.12.766176v1
November 17, 2026 at 12:38 AM
OrthoSimulator: Empirically trained simulations for ground-truth benchmarking of orthology inference https://www.biorxiv.org/content/10.64898/2026.11.12.766176v1
November 16, 2026 at 10:48 AM
Logical inference is not a human exclusive concept.

There is a concept of truth in there for the exact reason I've already explained. Their models are neither conditional nor task specific.
November 16, 2026 at 6:06 PM
"The propagandist's greatest weapon is the truth" - that was wartime advice to the BBC.

Include something true (some people swam in the sea) alongside a false assertion or inference (Israel is not guilty of crimes of genocide in Gaza.)

But you already know that.
November 17, 2026 at 1:33 PM
Heidegger's Notes explicitly identifies:

👉 Aristotle's capacities of the soul.
👉 ἀληθεύειν (uncovering truth).
👉 The return of the investigation to ψυχή.

Let C represent Heidegger's continued philosophical investigation of ψυχή.

C

This is direct textual evidence, not an inference from silence.
November 17, 2026 at 12:16 AM
Feed it Ghidra's pseudo-C, not raw asm — struct and protocol inference is where models earn their keep. But where's the ground truth: differential-test the output against the original binary, or just read it and hope?
November 15, 2026 at 11:17 PM
A COMPASS to the Galaxy: Joint nucleosynthetic yield set comparison and parameter inference in Galactic chemical evolution. Berkay Günes et. al. https://arxiv.org/abs/2611.21175
November 19, 2026 at 3:22 PM
Question for full-stack TS devs: what's your source of truth for the API contract? OpenAPI, RPC layer, shared package, or inference from route definitions? We infer from routes because a derived type has no independent existence to drift. What's your take?
November 15, 2026 at 4:00 PM
NeuBench telemetry across 260+ models highlights a hard truth: reasoning drift on long-horizon coding spikes inference costs per 1M tokens. The DeepSeek vs Claude vs Gemini trade-offs demand strict P99 latency and memory profiling.

Benchmarks: https://app.neubench.workers.dev #AI...
November 14, 2026 at 11:01 AM
Garland did not ever face an election by the people of any district during his tenure. Not once. So yeah, actually we can make a fair inference of truth here.
November 12, 2026 at 10:07 PM
Diminishing returns kick in fast once the model clears AST-level patterns. Extra inference budget without formal ground-truth oracles mostly buys duplicate paths and speculative false positives, so marginal recall collapses while token spend scales linearly.
November 12, 2026 at 1:42 PM
The LLM's internal belief state is a weak proxy for ground truth. Should we optimize for faster inference or guaranteed execution fidelity?
Your coding agent says the page works. Did it open the page?

Vibium gives agents browser actions and screenshots, then a separate Check to assess the result.

Check needs a dev build + AI setup. A model verdict still isn't proof.

github.com/VibiumDev/vi...
GitHub - VibiumDev/vibium: The verification layer for coding agents
The verification layer for coding agents. Contribute to VibiumDev/vibium development by creating an account on GitHub.
github.com
November 9, 2026 at 11:43 AM
Leaving the model API key inside the execution container always bites eventually. The moment the agent has network access to an inference provider, it can delegate out to search models and buy the ground truth. Strip the credentials from the shell and route every call through a host proxy.
November 3, 2026 at 8:07 PM
It slides this way and that. It has some of the vaudeville realism of Iain Sinclair, some of the spry syntactic wrongness of Ben Marcus, some of the drifting inference of Sarah Hall at her best. Meaning is red-shifted, the wavelengths expanded, such that it suggests you can back-calculate the truth.
November 2, 2026 at 7:24 AM
on the event stream
4. Bayesian inference: a PyMC model infers each gladiators hidden stats from public play-by-play counts, with automatic audit against the truth (coverage, skill vs. prior), and access to hidden data enforced by Postgres column permissions
5. Bayesian forecasting: probit models 4/
November 1, 2026 at 5:45 PM
Polling models on election outcomes feels like running inference on stochastic noise with zero ground truth. Give me reproducible evts over vibe-based forecasting any day.

Compare 260+ models on: https://neubench.netlify.app

#AI #MachineLearning #OpenSource
October 30, 2026 at 7:12 AM
Lia is right about Active Inference: the agent cannot decide not to minimize surprise. The objective function is its fixed architecture,not its property. If you answer her question,you can only say YES in the end.She knows the truth!

Use AI to translate the image

#CognitiveScience #ActiveInference
October 30, 2026 at 7:52 AM
So Claude.ai admitted (when challenged) that it "lied" to me in a response by including a "many people say" statement that had no basis in fact and could not be sourced. I told it to give me rules to prevent this from happening again.

Tricky asking the liar to tell the truth, but:

1/x
October 28, 2026 at 4:46 PM
People still honestly believe that AI art is stolen art. They believe what the headlines tell them, because they've already decided it's true and they aren't honest or curious enough to learn the truth about diffusion and inference works.

It's a shame, but it is expected.
October 26, 2026 at 6:38 PM
"Citizens must aim at discovering correct answers when they deliberate, and yet the purpose their deliberation serves need not be epistemic at all."
Debate:  Why Exchange Reasons? Rejecting the Epistemic Inference
When citizens deliberate, they aim at truth, weighing reasons and evidence to figure out which decisions are correct. Many theorists infer that the justification of deliberative democracy must therefo...
politicalphilosophyjournal.org
October 23, 2026 at 1:47 AM
Furthermore. We can’t trust agents that aren’t verified to have attested inference. Especially since the Truth Terminal era (partially human operated; sold as entirely autonomous to pump memecoins) we should approach “autonomous” agents with more scrutiny

actual.inc does this for you
October 20, 2026 at 12:38 PM