ARDD 2026 talk description: AI aging clocks, model benchmark and drug-discovery costs
Thumbnail from YouTube. Playing loads the YouTube player.
History & recommendationsOff
Keep recently viewed content on this browser. Your history stays on this device and is never sent to Harbor. Videos you play can also help order related video recommendations.
Open saved items & history →At a glance
Automatic summary · not editorially reviewedThe description attributes to Alex Zhavoronkov a talk on aging-first drug discovery: aging clocks evolving toward multi-omic foundation models, a benchmark where smaller fine-tuned models beat large language models on biological tasks, and AI validated only by repeated drug candidates. [1]
The description says aging clocks evolved from early epigenetic clocks to multi-omic foundation models such as Insilico's Precious GPT series. [1]
In a benchmark of 25,000 prompts across 18 frontier models, smaller fine-tuned models outperformed large language models on biological tasks. [1]
The description states AI is validated only when it repeatedly yields successful drug candidates, with about 5 million US dollars estimated to reach a developmental candidate. [1]
Aging clocks toward multi-omic foundation models
According to the description, Zhavoronkov presented at the 13th Aging Research and Drug Discovery Meeting on aging-first drug discovery. He described aging clocks as having evolved from early epigenetic clocks, likened to mechanical watches, to multi-omic foundation models such as Insilico's Precious GPT series. The description gives no details on how these clocks were built or validated. [1]
Benchmark: fine-tuned models versus frontier LLMs
The description reports a benchmark of 25,000 prompts across 18 frontier models in which smaller fine-tuned models outperformed large language models on biological tasks. It does not name the tasks, the models or the scoring. As a future direction it mentions large orchestrator models coordinating many specialized agents for continuous research. [1]
Validation claim and estimated candidate cost
The description states that AI is validated only when it repeatedly yields successful drug candidates, with about 5 million US dollars estimated to reach a developmental candidate. It does not say whether such candidates exist or how the estimate was derived. The meeting description also notes ARDD's North American debut in Boston on October 1–3, 2026, and its founding by Zhavoronkov in 2014. [1]
Context & limits
This is only the video description; the video and transcript were not reviewed. All claims are attributed to the description or creator. The benchmark, cost estimate and validation statement lack methods, model names and outcome data, so they cannot be independently checked here.
Based on the creator’s description; the full video and transcript were not analyzed.
Quoted passages in the original language
Aging clocks have evolved from “mechanical watches”
smaller fine-tuned models outperformed large LLMs on
AI is validated only when it repeatedly
About this video
Creator description · original textText language: English
ARDD · 29 min · Source published: 2026-10-22
Added to Harbor: 10/23/2026
Original on YouTube ↗Automatically categorized source. Creator statements are not an independent evidence review or personal medical guidance.