Benjamin Han

@BenjaminHan@sigmoid.social
682 Followers
229 Following
899 Posts

Husband, father, runner, German learner, piano player. A curious soul living in #PacificNorthwest. Working on Knowledge + #ML + #AgenticAI at .

#Running 5/25/18-7/13/25: (dist # time pace/mi date)

5K 774 21:11 6’49” 7/7/25
10K 240 44:16 7’07” 3/23/25
15K 17 1:09:25 7’27” 4/6/25
HM 59 1:39:07 7’34” 3/16/25
M 26 3:25:52 7’51” 4/13/25
50K 8 4:47:35 9’15” 6/22/25 (moving time)

2025: 1,558.8/2,400mi (2024: 2,375.4mi)
Max dist: 35.28mi 6/22/25

#nlp #nlproc #knowledgeGraphs #classicalMusic

Running recapshttps://www.youtube.com/playlist?list=PLz7qd_EMlRkR4hTs6HgmqK_LMgALFvemt
@randomized Job well done, body!

OK I am back home. Yesterday hike was just the perfect balance between hard, beautiful, challenging. I fear i won't find a match soon.

Think I've done my longest EG gain in 1 shot, from 1000 to 3000m .

Spent the night in a refuge and went backdown in the valley

Brain and myself very grateful to body for supporting us in those epic trips without a single blister ,black nail or painful muscle.

#vacations
#trailrunning
#runnersofmastodon
#running

@vaishakbelle Looking forward to preprint!

@vojtechcahlik We also explored grounding generation with reasoning in our “think while you write” paper: https://arxiv.org/abs/2311.09467

#genAi #nlp #nlproc #llm #reasoning

Think While You Write: Hypothesis Verification Promotes Faithful Knowledge-to-Text Generation

Knowledge-to-text generators often struggle to faithfully generate descriptions for the input facts: they may produce hallucinations that contradict the input, or describe facts not present in the input. To reduce hallucinations, we propose a decoding-only method, TWEAK (Think While Effectively Articulating Knowledge), which can be integrated with any generator without retraining. TWEAK treats the generated sequences at each decoding step and its future sequences as hypotheses, and ranks each generation candidate based on the extent to which their hypotheses are supported by the input facts using a Hypothesis Verification Model (HVM). We first demonstrate the effectiveness of TWEAK by using a Natural Language Inference (NLI) model as the HVM and report improved faithfulness with a minimal impact on the quality. We then replace the NLI model with a task-specific HVM trained with a first-of-a-kind dataset, FATE (Fact-Aligned Textual Entailment), which pairs input facts with their original and perturbed descriptions. We test TWEAK with two generators, and the best TWEAK variants improve on average for the two models by 2.24/7.17 points in faithfulness (FactKB) in in/out-of-distribution evaluations, respectively, and with only a 0.14/0.32-point decline in quality (BERTScore).

arXiv.org

I'm happy to speculate that our general technique for grounding explanations in LLM reasoning, presented at last week's XAI 2025 conference, could pave the way for finally cracking natural language explanations.
https://arxiv.org/abs/2503.11248

#AI #genAI #LLM #LLMs #ExplainableAI #AIsafety #NLP #ML

Reasoning-Grounded Natural Language Explanations for Language Models

We propose a large language model explainability technique for obtaining faithful natural language explanations by grounding the explanations in a reasoning process. When converted to a sequence of tokens, the outputs of the reasoning process can become part of the model context and later be decoded to natural language as the model produces either the final answer or the explanation. To improve the faithfulness of the explanations, we propose to use a joint predict-explain approach, in which the answers and explanations are inferred directly from the reasoning sequence, without the explanations being dependent on the answers and vice versa. We demonstrate the plausibility of the proposed technique by achieving a high alignment between answers and explanations in several problem domains, observing that language models often simply copy the partial decisions from the reasoning sequence into the final answers or explanations. Furthermore, we show that the proposed use of reasoning can also improve the quality of the answers.

arXiv.org

“On downhills, which you’re more likely to encounter regularly on a trail, your quad muscles lengthen more than they would on a flat or uphill, putting more tension on the muscle fibres. That eccentric nature of downhill running induces more lower limb muscle damage for up to several days after exercise, a 2020 scientific review published in Sports Medicine found.”

#running #trailrunning https://mastodon.nl/@Johan_Barelds/114850778797295261

Johan Barelds 🇪🇺 (@Johan_Barelds@mastodon.nl)

Nice read about the differences between Road and Trailrunning. "you want to move your feet as rapidly as you can tippy tap them while still maintaining good posture"" "The best downhill runners are moving their feet fast enough that any misstep is already corrected for by the next step." This is something I recognize already from when we where hiking the mountains decades ago. #Trailrunning #runnersofmastodon #UltraRunner #Running https://www.runnersworld.com/uk/training/motivation/a65316745/trail-running-versus-road-running/?utm_source=substack&utm_medium=email

Mastodon.nl door Stichting Activityclub
@Johan_Barelds Well executed! Congratulations!

Why isn’t there a super-automatic espresso machine that you can wear as a backpack or helmet

TECH INDUSTRY WTF ARE YOU EVEN DOING

@melsdung @Mr_GHARice I’ve found Parkrun to be a perfect motivator to run regularly!
@Dave3307 Welcome to the neighborhood!
×

@mdreid

Looks like a Virga Snowbow.