WEBVTT

00:00:00.000 --> 00:00:03.576
Who does an AI tool serve once it starts showing paid messages next to its answers?

00:00:03.576 --> 00:00:07.153
Today's news is about how the builders of these tools plan to earn money.

00:00:07.153 --> 00:00:10.729
That shift reaches into pharma work.

00:00:10.729 --> 00:00:14.305
As the seller's interests grow stronger, the basis for an answer may become harder

00:00:14.305 --> 00:00:17.882
to see from outside.

00:00:17.882 --> 00:00:21.458
In our field, we are the ones who must explain the basis for every claim and every

00:00:21.458 --> 00:00:25.035
figure.

00:00:25.035 --> 00:00:28.611
We cannot publish a machine's draft as it stands.

00:00:28.611 --> 00:00:32.187
So who should check that basis?

00:00:32.187 --> 00:00:35.764
The builders, the state, or the users?

00:00:35.764 --> 00:00:39.340
Let's keep that question in mind to the end.

00:00:39.340 --> 00:00:43.734
Our story begins with a builder changing the way it earns.

00:00:43.734 --> 00:00:48.129
When the source of income changes, so does the question of whom the tool tries to

00:00:48.129 --> 00:00:52.523
please.Here is how it works.

00:00:52.523 --> 00:00:56.918
A user asks for a picture, and a seller's message sits right beside the finished

00:00:56.918 --> 00:01:01.312
picture.

00:01:01.312 --> 00:01:05.707
What someone asks for hints at which message might work on them.

00:01:05.707 --> 00:01:10.101
The builder also says it wants to count how well each message works.

00:01:10.101 --> 00:01:14.496
So user behavior itself becomes a valuable record for sellers.

00:01:14.496 --> 00:01:18.890
Users are no longer the only ones paying.

00:01:18.890 --> 00:01:23.284
And a tool tends to serve whoever pays for it.Pharma staff ask these tools for help

00:01:23.284 --> 00:01:27.679
every day.

00:01:27.679 --> 00:01:32.073
How far could those requests feed someone else's business?

00:01:32.073 --> 00:01:36.468
Before adoption, I would check the contract terms and how inputs are handled.

00:01:36.468 --> 00:01:40.862
Staff also need to tell a neutral answer from a paid message on screen.

00:01:40.862 --> 00:01:45.257
That matters even more for free versions used at work.

00:01:45.257 --> 00:01:49.651
Write these rules into company policy, and people on the ground won't have to guess.

00:01:49.651 --> 00:01:54.046
Earning from paid messages is not a need of this builder alone.

00:01:54.046 --> 00:01:58.440
In the same week, other companies also made news about money.

00:01:58.440 --> 00:02:02.618
Building these systems means paying huge computing bills, year after year.

00:02:02.618 --> 00:02:06.797
Who carries that cost, and how?

00:02:06.797 --> 00:02:10.975
That is the question here.One company is expected to sell its stock to the public,

00:02:10.975 --> 00:02:15.154
while another has pushed that step back.

00:02:15.154 --> 00:02:19.332
With the outlook unclear, views are split.

00:02:19.332 --> 00:02:23.511
Once listed, a company is watched from outside through its numbers every quarter.

00:02:23.511 --> 00:02:27.689
Meanwhile, what users pay also differs widely from one builder to the next,

00:02:27.689 --> 00:02:31.867
according to one comparison.

00:02:31.867 --> 00:02:36.046
To me, that spread shows that competition among builders has not yet settled.So when

00:02:36.046 --> 00:02:40.224
a drug company chooses a tool, it should look at the vendor's finances, not just

00:02:40.224 --> 00:02:44.403
performance.

00:02:44.403 --> 00:02:48.581
A change in strategy can suddenly change prices or what you are allowed to use.

00:02:48.581 --> 00:02:52.759
Avoid depending on a single vendor, and keep room to switch.

00:02:52.759 --> 00:02:56.938
That is how you stay ready for the long term.

00:02:56.938 --> 00:03:01.116
This matters most when you rebuild workflows around these tools.

00:03:01.116 --> 00:03:05.295
Decide before signing how you would handle a price rise or the end of a service.

00:03:05.295 --> 00:03:09.473
The more money talk comes into view, the more the builders' stance on safety is

00:03:09.473 --> 00:03:13.652
questioned.

00:03:13.652 --> 00:03:17.830
That question reached lawmakers, who called the builders in.

00:03:17.830 --> 00:03:22.346
Who decides what safe AI means, and where?

00:03:22.346 --> 00:03:26.862
There is no settled answer yet.

00:03:26.862 --> 00:03:31.378
That is why words spoken before lawmakers carry so much weight.Before lawmakers, one

00:03:31.378 --> 00:03:35.894
expert warned that the builders are creating a danger with their own hands.

00:03:35.894 --> 00:03:40.411
On that same day, a company known for caution was accused of spreading panic.

00:03:40.411 --> 00:03:44.927
The same caution looks too weak to some and too much to others.

00:03:44.927 --> 00:03:49.443
Until outside rules set a standard, judgments will keep swinging by viewpoint.

00:03:49.443 --> 00:03:53.959
Political winds could even change how builders behave.In pharma, national rules and

00:03:53.959 --> 00:03:58.475
formal review define what a safe medicine looks like.

00:03:58.475 --> 00:04:02.991
For AI, safety is still close to self-reporting by the builders.

00:04:02.991 --> 00:04:07.507
So users should not rely on builders' words alone.

00:04:07.507 --> 00:04:12.023
They need their own standards.

00:04:12.023 --> 00:04:16.539
Which tasks may use AI, and where must a person check?

00:04:16.539 --> 00:04:21.056
Put that line in writing inside the company.

00:04:21.056 --> 00:04:25.572
That document lets you explain your choices.

00:04:25.572 --> 00:04:30.088
Then, however outside opinion swings, your internal standard can stay put.

00:04:30.088 --> 00:04:34.604
While the safety debate goes on, the technical distance between nations is quietly

00:04:34.604 --> 00:04:39.120
closing as well.

00:04:39.120 --> 00:04:43.479
When the difference in performance narrows, the reasons for choosing one tool over

00:04:43.479 --> 00:04:47.838
another change too.A large language model reportedly ran fast on equipment you could

00:04:47.838 --> 00:04:52.197
keep at home, not in a company's big data center.

00:04:52.197 --> 00:04:56.556
Strong performance is becoming usable on your own machines, without an outside

00:04:56.556 --> 00:05:00.914
service.The distance between the two nations is also estimated at its smallest ever.

00:05:00.914 --> 00:05:05.273
One Chinese developer is said to be the main reason.That developer's new version

00:05:05.273 --> 00:05:09.632
also placed near the top of a monthly ranking.

00:05:09.632 --> 00:05:13.991
In response, an American company is preparing a rival that publishes its inner

00:05:13.991 --> 00:05:18.350
workings for anyone to run.

00:05:18.350 --> 00:05:22.709
I see a shift.

00:05:22.709 --> 00:05:27.068
We are moving from choosing tools by performance to choosing by where they run and

00:05:27.068 --> 00:05:31.427
who controls them.Pharma work involves a lot of data that must never leave the

00:05:31.427 --> 00:05:35.786
company.

00:05:35.786 --> 00:05:40.144
Patient information and unpublished trial results are examples.

00:05:40.144 --> 00:05:44.503
An AI that runs on local machines lets you handle such data without sending it out.

00:05:44.503 --> 00:05:48.862
But which country built the model still needs careful review against company rules.

00:05:48.862 --> 00:05:53.221
Now we move from technology to money closer to pharma.

00:05:53.221 --> 00:05:57.580
There, what we don't know stood out more than what we do.

00:05:57.580 --> 00:06:01.859
Money reportedly changed hands between two AI companies close to drug discovery.

00:06:01.859 --> 00:06:06.138
But the report goes no further than its headline.The wording suggests the sum is

00:06:06.138 --> 00:06:10.416
settled, not just an estimate.

00:06:10.416 --> 00:06:14.695
Yet we don't know who pays whom, or what the money is for.

00:06:14.695 --> 00:06:18.974
We also don't know whether it arrives at once, or in installments as the work hits

00:06:18.974 --> 00:06:23.253
each goal.

00:06:23.253 --> 00:06:27.532
From a headline alone, the weight of this deal can't be judged.A big number can hide

00:06:27.532 --> 00:06:31.811
the terms behind it.

00:06:31.811 --> 00:06:36.089
Paid upfront, the same amount is real value.

00:06:36.089 --> 00:06:40.368
Paid on milestones, it never comes if the goals are missed.

00:06:40.368 --> 00:06:44.647
Which one it is changes what this deal means.Pharma professionals need the same care

00:06:44.647 --> 00:06:48.926
when reading partnership news.

00:06:48.926 --> 00:06:53.205
Hold your judgment until the payment terms and the direction of money are clear.

00:06:53.205 --> 00:06:57.484
In internal reports, separate what is known from what is not.

00:06:57.484 --> 00:07:01.762
And one headline cannot tell you how money flows across the industry.

00:07:01.762 --> 00:07:06.041
Keeping the unknown apart from the known also came up, in another form, in research

00:07:06.041 --> 00:07:10.320
on judging AI answers.

00:07:10.320 --> 00:07:14.391
When developers train an AI, they sometimes let another AI judge its answers.

00:07:14.391 --> 00:07:18.462
If the judge is lenient, the trained model grows lenient too.This study targets one

00:07:18.462 --> 00:07:22.533
error.

00:07:22.533 --> 00:07:26.604
The judge rewards content that isn't actually in the answer.

00:07:26.604 --> 00:07:30.675
If you tweak the question slightly and the score stays put, the judge probably isn't

00:07:30.675 --> 00:07:34.746
reading.

00:07:34.746 --> 00:07:38.817
The team then updates the scoring standard itself as training goes on.

00:07:38.817 --> 00:07:42.888
The rules aren't fixed.

00:07:42.888 --> 00:07:46.959
They grow along with the model.On medical questions, the researchers report more

00:07:46.959 --> 00:07:51.031
correct answers than with the older approach.

00:07:51.031 --> 00:07:55.102
But the summary doesn't show how the gains differ by model.

00:07:55.102 --> 00:07:59.173
The work has not yet been peer reviewed.Pharma will see more cases of one AI

00:07:59.173 --> 00:08:03.244
checking another.

00:08:03.244 --> 00:08:07.315
Picture a reply to a medical inquiry.

00:08:07.315 --> 00:08:11.386
Does it mention patients who should not take the drug?

00:08:11.386 --> 00:08:15.457
If the judge gives credit for a warning that isn't there, a risky reply passes.

00:08:15.457 --> 00:08:19.528
People must still spot-check that the supporting text really exists.

00:08:19.528 --> 00:08:23.599
Can we trace the basis for an answer?

00:08:23.599 --> 00:08:27.670
The same question is now moving through everyday tools and rules.

00:08:27.670 --> 00:08:32.241
Can we show the origin of writing or pictures produced by AI?

00:08:32.241 --> 00:08:36.812
For anyone handling promotional materials, there is no way around this question.A

00:08:36.812 --> 00:08:41.384
writing tool now points to what its answers are based on.

00:08:41.384 --> 00:08:45.955
One builder has said how it will meet a European rule on showing where text comes

00:08:45.955 --> 00:08:50.526
from.

00:08:50.526 --> 00:08:55.097
Meanwhile, in a popular game, a mark left in an AI-made picture revealed that AI was

00:08:55.097 --> 00:08:59.668
used.

00:08:59.668 --> 00:09:04.239
Tools, rules and accidental discovery are all pushing the same way, toward showing

00:09:04.239 --> 00:09:08.811
origins.In pharma materials, you can't publish a claim or a figure without

00:09:08.811 --> 00:09:13.382
explaining its basis.

00:09:13.382 --> 00:09:17.953
Reviewers ask for that, one item at a time.

00:09:17.953 --> 00:09:22.524
If an AI draft comes with its references, checking takes less work.

00:09:22.524 --> 00:09:27.095
If an AI-made figure carries a hidden mark, its use can be traced later.

00:09:27.095 --> 00:09:31.666
So I recommend logging which steps used AI.

00:09:31.666 --> 00:09:36.238
Without that log, you have no way to check when reviewers ask.

00:09:36.238 --> 00:09:40.809
Behind these stories about records and origins, a few quieter items sat closer to

00:09:40.809 --> 00:09:45.380
the daily work of users.

00:09:45.380 --> 00:09:49.248
Some events that got little coverage still bear directly on the work of people who

00:09:49.248 --> 00:09:53.117
use AI.Someone sued, saying the name of an AI system clashes with a name they

00:09:53.117 --> 00:09:56.985
already own.

00:09:56.985 --> 00:10:00.854
A tool released for camera glasses left its secret access codes unprotected.

00:10:00.854 --> 00:10:04.722
And in a fight over using other people's writing to train AI, a higher court upheld

00:10:04.722 --> 00:10:08.591
the lower court's view.

00:10:08.591 --> 00:10:12.459
The headlines were small.

00:10:12.459 --> 00:10:16.327
But they touch the basics of using AI: names, access codes and training material.

00:10:16.327 --> 00:10:20.196
A small headline does not mean a small impact.

00:10:20.196 --> 00:10:24.064
Small stories often become big problems later.When a drug company brings in AI, it

00:10:24.064 --> 00:10:27.933
always compares performance and price.

00:10:27.933 --> 00:10:31.801
But name rights, key storage and the rights to training data often slip off the

00:10:31.801 --> 00:10:35.669
checklist.

00:10:35.669 --> 00:10:39.538
If a key leaks, outsiders can use AI under your company's account.

00:10:39.538 --> 00:10:43.406
Rulings on training data also affect who bears responsibility in the vendor

00:10:43.406 --> 00:10:47.275
contract.

00:10:47.275 --> 00:10:51.143
Add these checks to your adoption process.

00:10:51.143 --> 00:10:55.012
Key storage and rights checks both show one thing.

00:10:55.012 --> 00:10:58.880
The job of verifying has already moved to the users.

00:10:58.880 --> 00:11:03.021
The more ways builders find to earn, the more the work of checking answers shifts to

00:11:03.021 --> 00:11:07.163
users.

00:11:07.163 --> 00:11:11.304
So far, neither governments nor lawmakers have taken on that job.

00:11:11.304 --> 00:11:15.446
It falls to us.

00:11:15.446 --> 00:11:19.587
We record origins, have people check the judges, and wait until the terms are clear.

00:11:19.587 --> 00:11:23.729
Next, we will see how the builders described their own rules to lawmakers.

00:11:23.729 --> 00:11:27.870
That account will help us set our own standards.
