Google’s Ex-CEO Drops FREE Superintelligent AI Agents 🤯 Smarter Than PhDs! | AI Revolution - video Dailymotion

Ai Revolution

🔥 BREAKING AI NEWS! Google’s former CEO has just unveiled FREE Superintelligent AI Agents that are outperforming PhDs! 😱🧠  In this AI Revolution episode, we cover: 🚀 The shocking capabilities of these AI agents 🎓 How they outperform PhD-level experts in multiple fields 💻 Real-world use cases and tools now available 🌐 What this means for the future of education, coding, and productivity 📉 Are traditional jobs and degrees at risk?  This could be one of the biggest AI releases of 2025 — and it's FREE to use! Tap in to see how it works and what it means for you.  🔔 Don’t forget to LIKE, COMMENT, and SUBSCRIBE for the most important AI updates!  #AIRevolution #GoogleAI #AIUpdate #SuperintelligentAI #AIvsPhD #FreeAI #OpenAI #ArtificialIntelligence #AItools #AI2025 #NextGenAI #AIforEveryone #MachineLearning #TechNews #AIBreakthrough #FutureOfWork #AIExplained #EmergingTech #ProductivityAI #AIvsHumans

Transcript

00:00FutureHouse just dropped four super-intelligent AI agents, and they're free to use right now.

00:08Backed by Eric Schmidt, the former Google CEO, this isn't some closed beta.

00:12It's public, powerful, and already outpacing PhDs in key tasks.

00:17Named Crow, Falcon, Owl, and Phoenix, each agent handles a different part of the workflow,

00:22from answering complex questions to designing brand new molecules,

00:25cutting down work that used to take weeks into minutes.

00:28So let's talk about it.

00:30Alright, let's start with the vibe of the platform.

00:32FutureHouse claims each agent is specialized from the ground up for science, not chit-chat.

00:38Crow is the quick-draw generalist.

00:40Ask it a technical question, and bam, it sifts through open-access papers and spits out a concise, citation-studded answer.

00:47Falcon goes deeper, slurping in dozens or even hundreds of full-text articles,

00:52plus proprietary databases like OpenTargets, then weaving them into long-form review.

00:56Owl, rebranded, from a prototype called HasAnyone, works like a detective,

01:02checking if someone's already done that crazy experiment so you don't waste six months reinventing the pipette.

01:08And Phoenix, still labeled experimental, is the chemistry brainchild.

01:12It proposes fresh compounds, predicts reactions, and even cost checks whether it's cheaper to buy or synthesize a molecule.

01:19All four can be chained together, which FutureHouse says lets a single researcher juggle workloads that used to require entire teams.

01:28Now, none of this happens in a vacuum.

01:30Google's own AI co-scienced announcement back in February lit a fire, and remember Google's 2023 genome system?

01:38It bragged about 40 supposedly new materials, but a later analysis showed none were genuinely novel.

01:45That flop still haunts the field, highlighting how hallucinations and shaky reasoning can torpedo flashy demos.

01:53FutureHouse is clearly trying to avoid that trap.

01:56In every blog post, they hammer on transparent reasoning and multi-stage evidence gathering.

02:02You can literally click through each step, watch the agent pick search terms, rank journals by citation graphs,

02:09and see why, say, nature genetics made the cut, while someone's bio-archive preprint did not.

02:16In one demo, Falcon pulled 32 papers, decided 24 were truly relevant,

02:21then distilled 62 separate chunks of evidence into a final answer.

02:26And you, the user, could inspect every breadcrumb.

02:30That visibility matters because, let's be real, scientists are a skeptical bunch.

02:35Even Sam Rodriguez, Andrew White, and the rest of the FutureHouse leadership admit

02:39today's large language models melt down on high-precision tasks.

02:43LLMs can't even count nitrogens in a molecule reliably.

02:47So Phoenix leans on external chemistry tools, patent databases, solubility predictors,

02:53reaction simulators, basically calculators stapled onto an LLM brain.

02:56The team runs its own wet lab in San Francisco so they can close the loop,

03:01generate a hypothesis, synthesize compounds, test them, feed the data back, tweak the model, repeat.

03:06FutureHouse calls this their four-layer architecture.

03:09At the base are generic AI tools.

03:12Think AlphaFold or, yes, your favorite gradient descent wizardry.

03:17Above that sit assistants like Crow and Falcon that execute specific workflows such as protein annotate.

03:23One rung higher is the future AI scientist that will design experiments end-to-end.

03:30And hovering over everything is the human researcher, the quest giver.

03:34Steering the big questions like curing Alzheimer's or, in today's flagship demo,

03:39tackling polycystic ovary syndrome.

03:42Alright, PCOS.

03:43This example is how FutureHouse tries to prove the platform isn't vaporware.

03:48Michaela Hinks, who leads their science team, says she came in cold.

03:52No PCOS background, just genetics chops.

03:55Step 1.

03:56She fired up Falcon and asked for a comprehensive breakdown of PCOS definitions,

04:01symptoms, diagnostic criteria, underlying causes, the whole deal.

04:05The agent ran a torrent of queries, hoovered up full-text studies and clinical trial records,

04:10filtered duplicates, and produced an overview in minutes.

04:14The transparent trace showed not just titles and DOIs, but quality scores tied to citation networks.

04:20Then, Michaela went surgical.

04:22She switched to Crow and asked which genes keep popping up in PCOS genome-wide association studies.

04:29Crow rattled off big hitters, including the now infamous D&D1A,

04:33and linked each to multiple GWAS papers.

04:36But listing genes is table stakes, so next she pinged Owl.

04:41Has anyone used CRISPR screens to connect these GWAS hits to hyperandrogenism?

04:46Owl confirmed that, yeah, one group had shown overexpression of D&ND1A,

04:51boosting testosterone in vitro.

04:54However, nobody had nailed the mechanism.

04:56Boom, research gap.

04:58Identified in four short prompts instead of a weekend library crawl.

05:01Still curious, Michaela probed, do we know why increased D&D1A jacks up androgen levels?

05:09And the agents came up empty.

05:11That's the moment a wet lab experiment becomes justified.

05:14Future House argues most of us burn days or weeks to reach that point.

05:19Enter Phoenix.

05:21Given the open question, Michaela asked Phoenix to propose three novel,

05:25potentially drug-like compounds that might tamp down D&D1A-driven hyperandrogenism.

05:30Because no FDA-approved binders exist for the protein,

05:34Phoenix first mapped protein interaction partners, then fetched molecules that hit those partners.

05:40It checked each candidate's patent status, solubility, functional groups,

05:44even synthetic routes, tossing out anything derivative.

05:47By the end, it produced a little dossier for each molecule.

05:50Why it might modulate D&D1A, how much it would cost to procure,

05:54and whether you'd need to tweak it into a pro-drug down the line.

05:58Future House says tasks like patent searching or log-S prediction rely on specialized tools.

06:03Because again, pure LMs are comically bad at chemistry arithmetic.

06:09Now, metrics.

06:10Future House ran benchmarks, lit QA precision and accuracy,

06:14head-to-head against frontier search models such as XAI plus human PhD literature sleuths.

06:21On retrieval precision, correct answers over answered questions,

06:24and accuracy, correct over all questions, their agents topped the charts.

06:29They haven't published raw percentages in the press articles, but the claim is superhuman.

06:34To be fair, the blog concedes Phoenix is less battle-tested, so it may make mistakes,

06:39and is out there to gather public feedback.

06:42Rapid iteration is the strategy.

06:44Nobody's pretending perfection.

06:47That honesty matters, especially after the G-nome disappointment

06:51and the broader worry that hallucinations could slip unvetted claims into real experiments.

06:57Gale is the other buzzword.

06:59PubMed holds 38 million papers.

07:01ClinicalTrials.gov tracks more than 500,000 trials,

07:05and scientists juggle thousands of siloed tools.

07:08Future House argues individual labs can't scrape, store, or rate limit that much data.

07:13They lack the engineering muscle.

07:15So the nonprofit wants to be the infrastructure layer.

07:18API calls so your screening pipeline can auto-kick literature reviews

07:22or contradiction hunts every night at 2 a.m.

07:24Everything is free, publicly available, and open to feedback,

07:27with the source code for key components opened under permissive licenses.

07:32The board includes names like Scientist Entrepreneur, Adam Marblestone,

07:36and Schmidt's backing presumably keeps the lights on without VC pressure to monetize everything tomorrow.

07:42Of course, skepticism hasn't vanished.

07:44The TechCrunch write-up notes that so far, Future House hasn't produced an entirely novel discovery.

07:50No new alloy, no first-in-class drug, despite fancy agents.

07:53The nonprofit counters that even Google's G-nome, with its 2023 fanfare,

07:59couldn't deliver net new materials,

08:01so the bottleneck is partly experimental logistics, not just algorithm smarts.

08:05They run a physical lab precisely to cross that last mile.

08:09Still, until a peer-reviewed breakthrough pops out,

08:12the broader scientific community will stay cautiously optimistic at best.

08:16Future House's launch timing is intriguing, too.

08:19The AI sector is flush with capital.

08:21OpenAI and Anthropic both publicly brag about accelerating science.

08:26Yet many bench scientists complain that general-purpose chatbots are unreliable,

08:30hallucinate, or sight-retracted studies.

08:33Future House tries to thread that needle by restricting its agents to carefully curated corpora,

08:38weighing source provenance and exposing every reasoning step.

08:42In demos, you can even watch the agent ditch low-impact journals

08:46in favor of higher citation density or stronger experimental methods.

08:50It's nerdy, sure, but it might be the transparency researchers need to trust an AI collaborator.

08:56Now, about that decade-long roadmap.

08:58Future House envisions the full AI scientists as an autonomous system

09:02that can read every paper in biology, spot contradictions,

09:06devise hypotheses, design and execute experiments,

09:10evaluate the data, and iterate,

09:12basically compressing the scientific method.

09:15Humans remain in charge of the quest,

09:17the top-level framing of what matters.

09:19Whether that's curing Alzheimer's, decoding the brain,

09:23or universal gene delivery,

09:24the idea is to let machines handle the brute-force literature grinding

09:28and high-dimensional search,

09:30while humans judge significance and ethics.

09:33If it works, the company says,

09:34we could clear today's global backlog of unexplored opportunities,

09:39thousands of promising leads buried in PDF supplemental sections,

09:43simply because no one has time to compare them.

09:45All of this lands in a world already buzzing about AI ethics.

09:49Future House's insistence on open-sourcing and auditable reasoning

09:52is meant to address risk, but unintended consequences,

09:56say, a hallucinated synthesis route

09:58that actually produces a toxic byproduct, still loom.

10:01The nonprofit is betting that transparency

10:04plus community feedback beats closed-door secrecy.

10:08They've even plastered a big

10:09please-provide-feedback banner over Phoenix's chemistry tools.

10:13So what's next?

10:14Short-term, Future House wants to widen the platform's range

10:17in just raw lab data, not just published papers.

10:20Picture an agent that sees your single-cell RNA sequence results

10:24in real time and immediately correlates them

10:26against millions of past experiments.

10:28The team thinks that within a couple of years,

10:30most discoveries will carry an AI-driven component,

10:34whether at the literature stage,

10:35the analytical stage, or both.

10:37Long-term, if they manage to keep hallucinations,

10:40bias, and experimental error in check,

10:43they might really bend the curve of scientific productivity,

10:46which economists have shown has stagnated

10:48even as paper counts soar.

10:51And look, hype aside, the launch day invitation is simple.

10:56Go to platform.heo.io, sign up for free,

10:59query crow, fire falcon, let owl double-check your idea,

11:03then ask Phoenix if it can cook you a molecule.

11:06Whether you're in academia with a dropbox

11:08full of half-read PDFs or a biotech startup

11:11drowning in assay data,

11:13the promise is a turbo button you didn't have yesterday.

11:17Will it deliver?

11:18History says skepticism is healthy.

11:20But the alternative?

11:22Wading through 38 million PubMed articles,

11:24500,000 clinical trials,

11:26and who knows how many supplementary spreadsheets by hand,

11:29feels, frankly, impossible.

11:31If Future House's agents can shave even 10% off that grunt work

11:35without injecting hallucinated nonsense,

11:37that alone could make them worth the hype.

11:40And if, as the founders hope,

11:42an autonomous AI scientist emerges before 2035,

11:47well, we may be telling our grad students stories

11:49about the Dark Ages

11:50when we had to parse every p-value ourselves.

11:54So, are we really about to let AI outperform top researchers

11:58and take over the front lines of discovery?

12:01Let me know what you think in the comments.

12:03Thanks for watching,

12:03and I'll catch you in the next one.

12:05See you in the next one.

Google’s Ex-CEO Drops FREE Superintelligent AI Agents 🤯 Smarter Than PhDs! | AI Revolution

Category

Transcript

Recommended