Today’s three news stories,
happen in:
a laboratory.
a creator’s workspace.
and an everyday workplace training setting.
Though seemingly different,
they all answer the same question:
Once AI can do the work, what exactly is the role of humans going forward?
Scientists are responsible for:
verification.
Creators are responsible for:
selection.
And everyday workers must now learn:
how to work alongside AI.
① Claude Runs 21 Hours Processing 210 Million Tokens, Narrows 200,000+ Enzymes to Discover a New ART System
The first story is very special.
On September 23, Anthropic announced:
the first results from its newly formed:
Life Sciences Research Group.
This time it’s not about:
Claude summarizing papers,
or writing reports for scientists,
but about having many Claude Agents
actively participate in:
scientific discovery.
About 950 Claude Agents Searching for Answers Simultaneously
Anthropic explained that their research team built a system called:
Large-scale Agentic Search,
where approximately
950 Claude Agents
worked within
21 hours,
processing over
210 million tokens.
Their task was to explore
large amounts of DNA and protein data
to find potential Reverse Transcriptase systems
worthy of real laboratory research.
Starting with Over 200,000 Candidates
The team first gathered over
200,000 Reverse Transcriptase sequences.
Manually reviewing these would be practically impossible.
Claude Agents began by:
locating nearby genes,
comparing sequences,
analyzing structures,
and reading relevant literature.
They judged which combinations were not already known systems.
The candidate pool was then reduced to roughly
3,500 systems,
and eventually down to
20.
This exemplifies AI's most suitable scientific role:
not declaring,
"I've found the truth,"
but rapidly narrowing an otherwise unmanageable
search space.
The Final Discovery: ART
Among the most notable candidates,
the researchers named it:
Array-associated Reverse Transcriptases,
or
ART.
It appears in bacteriophages,
viruses that infect bacteria.
The team observed a unique structure:
a Reverse Transcriptase,
adjacent to a partner gene,
beside a long array of repeated DNA sequences.
The arrangement reminded researchers of another well-known biological system:
CRISPR.
But It’s Not Yet Accurate to Call It “The Next CRISPR” Discovered by Claude
This point is crucial.
Anthropic themselves have not made such a claim.
What is currently known is that:
the Reverse Transcriptase itself was previously identified.
What the Claude Agents truly helped with was:
connecting scattered clues into a previously under-described
gene plus DNA array system.
In short,
Claude linked the scattered data into a coherent system worthy of research.
But what exactly does ART do?
What biological function does it serve?
Could it become a new biotechnology tool?
Those questions remain unanswered.
AI Finds Candidates; Humans Conduct Experiments
This highlights the most important part of the story.
Claude didn’t:
grow cells spontaneously in a lab,
operate lab equipment,
or announce experimental results.
Anthropic emphasized that:
all actual wet-lab experiments are still performed by
human scientists.
Follow-up experiments are ongoing
to confirm ART’s exact function.
This division of labor is quite clear:
AI:
finds possibilities.
Humans:
verify real-world truth.
Different from AI Helping Pharma Research with Data
On September 17, SasaDaily reported that:
Novo Nordisk started bringing Claude into
drug research.
There, AI mainly organized massive research data and helped scientists manage knowledge work.
Today’s Anthropic announcement goes a step further:
AI doesn’t just read what’s known,
but also helps researchers decide
which unknown candidates merit the next experiment.
If this model continues to succeed,
scientific research will be shortened not by the experiments themselves,
but by
greatly reducing the lengthy search before experiments.
But Verification Remains the Essential Role AI Can’t Replace in Science
A single agent could propose
1,000 hypotheses overnight.
A real lab cannot verify all overnight.
So as AI speeds up,
the next bottleneck may become:
experiment capacity,
personnel,
equipment,
materials,
repetitions, and
peer review.
In other words, AI makes
proposing answers
very cheap,
but
proving those answers true
is still costly.
This may become AI science’s most critical challenge.
② YouTube Integrates Gemini Directly Into Editing: Chat Your Way to Rearranging, Trimming, and Syncing to Music
The second story is more directly related to daily creators.
On September 23, YouTube announced at
Made on YouTube
a series of new Creator AI tools.
One highlight is that YouTube no longer just uses AI to:
generate an image,
or generate a video,
but brings Gemini into the
editing process.
Create a Draft, Then Chat With Gemini to Refine It
The new
Conversational Editing Assistant
will be part of
YouTube Shorts
and
YouTube Create.
Creators can start with a prompt
to generate a
first draft.
Instead of regenerating the whole video for edits,
they can use natural language commands to:
rearrange frames,
trim sections,
sync music to the beat,
and adjust pacing.
They can also return to the traditional timeline
for manual fine-tuning.
In essence,
AI does the initial work,
and creators continue to polish it.
This Is Very Different from “Prompt Once and Get the Perfect Video”
Previously, generative video often faced this problem:
The first version was about 80% correct,
but the remaining 20%
was wrong.
The only option was to regenerate, again and again.
YouTube wants to move the workflow closer to real editing software.
Not:
Prompt → Final Video,
but rather:
Prompt → Draft → Conversation → Manual Edit.
This aligns more closely with how
actual creators work.
YouTube Studio Will Also Enable A/B Testing with Up to 3 Versions of the Same Video
YouTube said that YouTube Studio
will add
video A/B testing.
Creators can prepare up to
3 different cuts
and compare
which version better retains viewers.
A/B testing was previously commonly applied to:
thumbnails,
titles,
but now it expands to
the video content itself.
This suggests that the next step for AI video tools may not be
“help you create the perfect version once,”
but rather
“produce multiple versions cheaply and let real audiences decide.”
A Crucial Emerging Issue: AI Can Mimic Your Face and Voice
As video generation becomes easier,
so does
impersonation.
YouTube already has
likeness detection
to help creators find
suspected AI-generated content using their own faces.
YouTube now says that later this year it will combine
speaking voice detection
with
facial detection.
This means,
not only assessing
“Does this face look like you?”
but also
“Is this voice also impersonating you?”
Likeness Detection Will Also Come to Mobile Apps
YouTube plans to bring
likeness detection
to its
mobile app.
This is significant because
managing AI deepfakes
is more accessible when creators can
handle it directly where they manage their channels daily,
not just sitting behind a desktop.
YouTube Is Also Testing AI That Learns Your Comment Moderation Style
Another relatively low-profile but practical feature is:
AI-powered comment moderation.
YouTube will allow some creators to opt in.
The AI will gradually learn
how this channel typically manages comments.
Instead of applying the same moderation rules for all,
it will understand
your community standards.
Like video generation,
AI is not just about
content creation,
but is entering
creator operations.
Live Streams Will Also Get Real-Time Auto Dubbing
YouTube also announced:
live streams will include
real-time auto dubbing.
This means
AI will perform cross-language processing
during the live stream itself,
unlike generating dubbing tracks after uploading videos.
If mature,
a creator could stream in one country
while viewers in other languages
watch simultaneously in their own language.
This will directly affect
whether creators can enter multiple markets.
YouTube Is No Longer Asking, “Should Creators Use AI?”
YouTube reports that by August 2026,
hundreds of thousands of channels
already use Gemini Omni’s creation capabilities daily.
Additionally, U.S. data shows that among 14-44 year olds
who uploaded videos in the past year,
72%
have used AI to assist
with creation or editing.
These numbers reflect YouTube’s perspective,
not global creators entirely,
but they clearly indicate
the platform no longer asks:
“Should creators use AI?”
but
“At which step of content creation should AI be integrated?”
③ Verizon Invests $70 Million: AI Training Transitions from Perk to Essential Workplace Skill
The third story involves no new models,
no new chips,
not even a new app,
but possibly a more direct impact on regular workers.
On September 23, Verizon announced:
Verizon AI Skills for America,
with a total investment of
$70 million.
The goal is to provide
free AI skills training.
Where Does the $70 Million Come From?
Verizon said that
$50 million
is new investment,
while
$20 million
comes from an existing
Reskilling and Career Transition Fund.
This means Verizon isn’t merely uploading free videos online,
but rather formally directing resources originally aimed at
workforce transition,
reemployment, and
skills upgrade
into AI training.
Who Is Eligible to Learn?
Target groups include:
job seekers,
those new to the workforce,
workers displaced by technical shifts,
educators,
small businesses,
and anyone seeking
to reskill.
Verizon is not focusing solely on teaching prompt engineering,
but also training on how to incorporate AI
into real work.
Training Is Not Provided by Verizon Alone
Companies contributing training content include:
IBM,
Google,
Anthropic,
Microsoft,
Coursera,
and OpenAI.
Additionally, Verizon will partner with
Goodwill Industries,
LISC,
and NACCE,
among other community and nonprofit groups,
to bring training directly to those in need of employment support.
Reuters reported Verizon aims to train
hundreds of thousands
through this program.
Why Is a Telecom Company Teaching AI?
Because AI skills
are evolving from:
a bonus for tech jobs
to:
a fundamental ability for regular jobs.
Just weeks ago,
UBS announced plans requiring some new hires
to prove AI fluency.
Verizon’s approach is at the opposite end:
not “Don’t come if you lack AI skills,”
but rather
“Let’s give more people the opportunity to learn first.”
Both Sides Should Be Viewed Together
If companies demand AI skills,
but good AI training is only accessible to
high-paid tech workers,
AI will likely widen the
skill gap.
Verizon noted that some premium AI training
can cost more than
$700 per person annually.
What Verizon aims to do is
remove that cost barrier,
giving more
job seekers,
small businesses, and
career changers
access to AI upskilling.
But “Taking an AI Course” Doesn’t Equate to True AI Capability
This is the key challenge ahead.
AI skills cannot end with
watching 10 hours of videos and
getting a certificate.
Truly valuable skills at work should enable you to answer:
Which steps in my job are suitable for AI?
Which cannot be delegated?
How do I verify outputs?
Can confidential information be included?
How do I detect AI errors?
Is time saved or results improved?
These are the hallmarks of
real AI fluency.
All Three News Items Are About Redefining Human Roles
In Claude’s scientific research:
AI quickly narrows 200,000 candidates down to 20,
but
humans must do the experiments.
In YouTube:
AI can help edit, arrange, and create multiple versions,
but
humans decide what’s worth publishing.
In Verizon’s case:
AI can take over more tasks,
but
humans must relearn how to work.
So as AI strengthens,
humans do not disappear.
Human work shifts layers—from:
searching all data,
to verifying the most important data;
from manually creating first drafts,
to judging which versions deserve to stay;
from knowing how to type or use Office,
to
knowing how to properly integrate AI into workflows.
This may be the true common thread among today’s three news stories:
AI is making
generating candidate answers
increasingly cheap.
Thus, the real value of humans concentrates on:
judgment,
verification,
responsibility,
and
knowing when not to blindly trust AI.
Today, make progress with AI a little each day.
Learn one AI skill daily.
Save a bit of time daily.
Improve your ability every day.
SasaDaily is here to grow with you.
