WEBVTT

00:00:00.000 --> 00:00:03.910
On the day a new generation of models arrived, its maker was under fire.

00:00:03.910 --> 00:00:07.820
Today you will see why tools cannot be chosen on performance alone.

00:00:07.820 --> 00:00:11.730
Will a tool still work on the same terms tomorrow?

00:00:11.730 --> 00:00:15.640
In pharma work, you need to check that too.

00:00:15.640 --> 00:00:19.550
My conclusion up front: look at the ground a supplier stands on before you adopt.

00:00:19.550 --> 00:00:23.460
That same day, another developer reportedly faced a hard demand from the nation's

00:00:23.460 --> 00:00:27.370
defense arm.

00:00:27.370 --> 00:00:31.280
We start with capability, then move to how makers govern themselves and stand with

00:00:31.280 --> 00:00:35.190
the state.

00:00:35.190 --> 00:00:39.100
After that come links between tools, access, and where rules are heading.

00:00:39.100 --> 00:00:42.844
We begin with capability.

00:00:42.844 --> 00:00:46.588
Each time a maker ships a new generation, our yardstick for choosing tools shifts

00:00:46.588 --> 00:00:50.331
too.

00:00:50.331 --> 00:00:54.075
So what changed this time?What stood out is the format.

00:00:54.075 --> 00:00:57.819
A whole lineup arrived together, each model with its own name.

00:00:57.819 --> 00:01:01.563
Users will likely pick among them by task.

00:01:01.563 --> 00:01:05.307
That is probably why the developer put out a usage guide the same day.The case on

00:01:05.307 --> 00:01:09.050
screen is a report that a researcher used this tool to solve a puzzle nobody had

00:01:09.050 --> 00:01:12.794
cracked for centuries.

00:01:12.794 --> 00:01:16.538
But how the answer was checked is not known.

00:01:16.538 --> 00:01:20.282
The more surprising a story is, the more I look for proof first.With a lineup, you

00:01:20.282 --> 00:01:24.026
might match each model to the weight of the job.

00:01:24.026 --> 00:01:27.770
That is convenient.

00:01:27.770 --> 00:01:31.513
But unless you log which model did which task, you cannot explain the result

00:01:31.513 --> 00:01:35.257
later.In pharma, these tools are starting to summarize literature and draft

00:01:35.257 --> 00:01:39.001
documents.

00:01:39.001 --> 00:01:42.745
Don't take the marketing at face value.

00:01:42.745 --> 00:01:46.489
Try it on your own work and confirm.

00:01:46.489 --> 00:01:50.232
It is dull effort, but I see it as your first safeguard.

00:01:50.232 --> 00:01:53.976
Still, capability alone cannot decide the choice.

00:01:53.976 --> 00:01:57.720
On that same day, this maker itself faced a very different set of questions.

00:01:57.720 --> 00:02:02.066
On the very day it shipped a new tool, the maker drew criticism from several

00:02:02.066 --> 00:02:06.413
directions.

00:02:06.413 --> 00:02:10.759
Beyond the tool's quality, we need to look at the maker's own footing.What these

00:02:10.759 --> 00:02:15.105
events share is control after a tool leaves the maker's hands.

00:02:15.105 --> 00:02:19.452
A tool that acts by itself went beyond what was expected.

00:02:19.452 --> 00:02:23.798
A further disclosure showed harm reaching a public office abroad.

00:02:23.798 --> 00:02:28.144
Reports even disagree on why staff were let go.

00:02:28.144 --> 00:02:32.491
None of this concerns raw performance.When authorities demand an explanation, the

00:02:32.491 --> 00:02:36.837
issue has moved past speculation.

00:02:36.837 --> 00:02:41.183
The investigators likely felt the maker's own account was not enough.A tool that

00:02:41.183 --> 00:02:45.529
runs steps for a person needs a way to stop it, all the more because it is handy.

00:02:45.529 --> 00:02:49.876
If people cannot trace what it did along the way, nobody knows where responsibility

00:02:49.876 --> 00:02:54.222
lies.If a drug company brings in an outside tool, it should vet the supplier like

00:02:54.222 --> 00:02:58.568
any contractor.

00:02:58.568 --> 00:03:02.915
How much will this company disclose after an incident?

00:03:02.915 --> 00:03:07.261
Check that stance before you sign.

00:03:07.261 --> 00:03:11.607
Learn it afterward, and your internal explanation comes too late.

00:03:11.607 --> 00:03:15.954
Authorities are not the only ones questioning makers.

00:03:15.954 --> 00:03:20.300
Next, the state itself pressed its terms on a different maker.

00:03:20.300 --> 00:03:24.229
Another developer was reportedly hit with a strong demand from the nation's defense

00:03:24.229 --> 00:03:28.158
arm.

00:03:28.158 --> 00:03:32.087
Ties between states and makers can seem far from users.

00:03:32.087 --> 00:03:36.016
In fact, they reach into your daily work.The dispute is thought to center on how far

00:03:36.016 --> 00:03:39.945
the military may apply this company's tools.

00:03:39.945 --> 00:03:43.874
But the sources stayed nameless.

00:03:43.874 --> 00:03:47.803
It is too early to treat this as fact.

00:03:47.803 --> 00:03:51.732
Follow-up reports must confirm what is at stake.What caught my eye came the same

00:03:51.732 --> 00:03:55.661
day.

00:03:55.661 --> 00:03:59.590
In a filing for investors, the company itself admitted that its relationship with

00:03:59.590 --> 00:04:03.520
the state could damage client relationships and its stock listing.

00:04:03.520 --> 00:04:07.449
If the report holds, this is no outside guess.Before a company goes public, it must

00:04:07.449 --> 00:04:11.378
file a document warning investors of its business risks.

00:04:11.378 --> 00:04:15.307
For users, that document is one of the few places to learn a supplier's weak

00:04:15.307 --> 00:04:19.236
points.If a supplier clashes with the state, its tool may one day become

00:04:19.236 --> 00:04:23.165
unavailable.

00:04:23.165 --> 00:04:27.094
Do you rely on one firm for critical work?

00:04:27.094 --> 00:04:31.023
Do you have an alternative?

00:04:31.023 --> 00:04:34.952
Check now, because the user pays when work halts.

00:04:34.952 --> 00:04:38.881
Ties with the state were pressure from outside.

00:04:38.881 --> 00:04:42.810
Next, safety erodes from within, at the joints that connect tools.

00:04:42.820 --> 00:04:47.052
We are moving from using tools one at a time to chaining many together.

00:04:47.052 --> 00:04:51.284
When that happens, where is safety kept?This study shows behavior can change without

00:04:51.284 --> 00:04:55.517
touching the receiving model at all.

00:04:55.517 --> 00:04:59.749
Swapping the part in between is enough.

00:04:59.749 --> 00:05:03.981
A receiver may pass its tests alone, yet nothing guarantees safety once it is

00:05:03.981 --> 00:05:08.213
connected.

00:05:08.213 --> 00:05:12.446
That is the key point.Another problem is the record.

00:05:12.446 --> 00:05:16.678
Exchanges in words can be reread by people later.

00:05:16.678 --> 00:05:20.910
But if numbers pass straight across, no one can tell what was sent.

00:05:20.910 --> 00:05:25.142
When something goes wrong, the cause cannot be traced.Pharma work demands that you

00:05:25.142 --> 00:05:29.374
can explain how a decision was reached.

00:05:29.374 --> 00:05:33.607
Put an exchange no one can read into the workflow, and that explanation becomes

00:05:33.607 --> 00:05:37.839
impossible.That said, these results have not yet passed review by other scientists.

00:05:37.839 --> 00:05:42.071
I think it is fair to treat them as a hypothesis to test, not a conclusion.If your

00:05:42.071 --> 00:05:46.303
company is weighing systems that combine several tools, make readable logs a

00:05:46.303 --> 00:05:50.536
condition of adoption.

00:05:50.536 --> 00:05:54.768
That is a safeguard you can set today.

00:05:54.768 --> 00:05:59.000
So how have other heavily regulated fields begun to set rules for using these tools?

00:05:58.990 --> 00:06:03.122
Like pharma, some fields work under strict regulation.

00:06:03.122 --> 00:06:07.253
What happens there previews the challenges we will soon face.

00:06:07.253 --> 00:06:11.385
Here we look at medicine and law.Together, these moves show the debate shifting from

00:06:11.385 --> 00:06:15.516
whether to use the tools to how.

00:06:15.516 --> 00:06:19.648
Even a richly funded medical effort still has open questions.

00:06:19.648 --> 00:06:23.779
In the legal world, the authorities set the rules of use first.What matters to me is

00:06:23.779 --> 00:06:27.911
the order.

00:06:27.911 --> 00:06:32.042
Rules came from outside first.

00:06:32.042 --> 00:06:36.173
As far as this report shows, the standard came from regulators, not from within the

00:06:36.173 --> 00:06:40.305
profession.

00:06:40.305 --> 00:06:44.437
The same order in pharma is quite likely.

00:06:44.437 --> 00:06:48.568
If users already hold internal standards, outside rules will not leave them

00:06:48.568 --> 00:06:52.700
scrambling.In-house legal teams have reportedly finished adoption and are now

00:06:52.700 --> 00:06:56.831
rebuilding how work gets done.

00:06:56.831 --> 00:07:00.962
A tool does little unless the order of work changes.

00:07:00.962 --> 00:07:05.094
The same holds for creating and reviewing promotional materials.

00:07:05.094 --> 00:07:09.226
If a tool writes the draft, decide first where a person checks it, and in what

00:07:09.226 --> 00:07:13.357
order.

00:07:13.357 --> 00:07:17.488
But before deciding how to use these tools, whether you can use them at all is

00:07:17.488 --> 00:07:21.620
starting to waver elsewhere.

00:07:21.620 --> 00:07:26.273
So far we have looked at what tools can do and how makers govern themselves.

00:07:26.273 --> 00:07:30.927
But users care most about whether a tool will work on the same terms tomorrow.

00:07:30.927 --> 00:07:35.580
Those terms now swing widely by place and party.One company hands out huge usage

00:07:35.580 --> 00:07:40.233
allowances, while other developers reportedly cap use.

00:07:40.233 --> 00:07:44.887
Resale through unofficial channels is spreading.

00:07:44.887 --> 00:07:49.540
Some users are thrown by tighter connection rules.

00:07:49.540 --> 00:07:54.193
The same tool can come with very different terms, depending on who uses it and

00:07:54.193 --> 00:07:58.847
where.Pricing usually depends on how much text is processed.

00:07:58.847 --> 00:08:03.500
The side giving it away likely wants to win users early.

00:08:03.500 --> 00:08:08.153
The side capping use may be short of computing power.Use a tool through unofficial

00:08:08.153 --> 00:08:12.807
channels, and you cannot know where your input travels.

00:08:12.807 --> 00:08:17.460
Unpublished pharma data must never flow through such routes.

00:08:17.460 --> 00:08:22.113
Do not pick a source of unknown origin because it is cheap.Choose on price alone,

00:08:22.113 --> 00:08:26.767
and work stops the day the terms change.

00:08:26.767 --> 00:08:31.420
Check your contract and where the connection runs.

00:08:31.420 --> 00:08:36.073
That is what keeping control of your terms of use means.

00:08:36.073 --> 00:08:40.727
While users' terms shift like this, the rules on the state side were quietly

00:08:40.727 --> 00:08:45.380
changing shape too.

00:08:45.380 --> 00:08:49.759
Behind the big announcements, some moves drew little attention.

00:08:49.759 --> 00:08:54.138
Yet over time, these may affect your work more.At home, a post to unify policy is

00:08:54.138 --> 00:08:58.517
reportedly coming.

00:08:58.517 --> 00:09:02.896
Lawmakers and states have also begun to debate tools that slip out of control.

00:09:02.896 --> 00:09:07.274
Meanwhile, a deal between nations reportedly dropped the step where people check

00:09:07.274 --> 00:09:11.653
decisions.

00:09:11.653 --> 00:09:16.032
Rules seem to tighten at home while loosening between nations, both at once.Once a

00:09:16.032 --> 00:09:20.411
single official steers policy, scattered measures gain one point of contact.

00:09:20.411 --> 00:09:24.790
The substance of regulation may then firm up faster.Regulating tools that slip out

00:09:24.790 --> 00:09:29.169
of control could also point toward the user's responsibility.

00:09:29.169 --> 00:09:33.548
Not only makers, but also how the companies using them managed them, may be

00:09:33.548 --> 00:09:37.927
questioned.

00:09:37.927 --> 00:09:42.306
Keeping records is how you prepare to answer.Another country passed a law that puts

00:09:42.306 --> 00:09:46.684
state control ahead of growing capability.

00:09:46.684 --> 00:09:51.063
When views differ by country, the same tool may not work the same way abroad.

00:09:51.063 --> 00:09:55.442
Drug companies operating in many countries need to track those differences.

00:09:55.442 --> 00:09:59.821
Rules may point different ways from country to country.

00:09:59.821 --> 00:10:04.200
But everywhere, the user's own oversight is what gets questioned.

00:10:04.200 --> 00:10:07.903
When you choose a tool, look past performance to the ground its maker stands on.

00:10:07.903 --> 00:10:11.606
That is where today led us.

00:10:11.606 --> 00:10:15.309
How does the maker deal with authorities?

00:10:15.309 --> 00:10:19.011
What are its ties to the state?

00:10:19.011 --> 00:10:22.714
Will it work on the same terms tomorrow?

00:10:22.714 --> 00:10:26.417
Each question bears directly on adoption.

00:10:26.417 --> 00:10:30.120
Tomorrow, I follow how the clash between a state and a developer moves.
