The day Google's search revenue disappointed the market── Two ways to earn from AI, AI-found flaws, research on edited videos
Download the video (MP4, 10 MB)
News that AI has grown smarter is no longer rare. Today's stories ask what comes after that. Who earns money from that power, and who checks it?One company posted strong results and still saw its shares sold off. The ability to find holes in software is reaching attackers as well as defenders. Reports of saved effort in pharma work keep coming. But we cannot yet tell whether those figures would hold inside your own company. People have also started asking what an AI learned from. Today I want to look past raw strength. I want to look at its price and its evidence.
01Alphabet shares plunge
Source WBUR ／ The Times ／ Tech Insider ／ The Tribune
We begin with one of the world's most profitable companies. Its total sales grew strongly, yet investors sold the stock. They seemed to care less about overall growth than about how its core business earns, and where that is heading.
| Item | Figure / detail |
|---|---|
| Alphabet revenue | Up 25% year on year; Google Cloud surged |
| Google Search revenue | $63.4B; below market expectations |
| AI Mode ads | Monetize less than before |
| Alphabet market value | Reported to have lost about $500B |
Revenue: The Tribune; search revenue: Tech Insider; market value: The Times
The real question was how well ads can pay on a screen where AI writes the answer. A user who is satisfied by the answer may click fewer ads. Usage can grow while each visit earns less. I read the sell-off as fear of that outlook.
Classic search listed where answers could be found and sent people elsewhere. Ads sat along that route. When AI writes the answer itself, the route disappears. So does the old place to put an ad.
The value that vanished was enormous. Worry about a changing revenue model outweighed the strong figures.
This matters to pharma teams too. When doctors or patients look for information, more searches will end with the AI's answer. The idea that people will reach your own website is weakening. Now is the time to rethink how your information travels.
So is advertising the only way to earn from AI? Another company has chosen a different path.
02Two ways to make money from AI
Source blog.google ／ CRN ／ TechCrunch ／ Benzinga
The same AI can cost users very differently, depending on who pays. Drug companies pay on both paths. Knowing which path a vendor takes is where price talks begin.
- Role
- CEO, Alphabet and Google
- Statement
- Google's AI is increasing search queries
- Move
- AI Mode ads expand to 10 countries incl. Canada, India
- Role
- CEO, Microsoft
- Statement
- Explained growing Copilot use and usage-based billing
- Move
- $10B in quarterly revenue from the OpenAI deal
Benzinga reported Microsoft aims to charge more as agents do more work, while Google sees opportunity in ads
One side places ads beside the AI's answers and is taking them abroad. The other charges for each piece of work done with its AI. It already earns large sums from a deal with a major model developer. The first earns from advertisers. The second earns directly from businesses. Neither is right or wrong. The difference is who pays.
When you pay per use, the bill rises as the AI works harder. The longer the tasks it takes on, the more it costs. There are reports of plans to raise prices in line with the workload.
If a pharma unit scales up AI, the first quote is not enough. Estimate the cost a year later, with more users and more tasks. Success itself can push you over budget. Set usage caps for each department, and review usage every month.
Beyond pricing, the power of AI itself is starting to change how systems are defended.
03Vulnerabilities found by AI
Source Mashable ／ Ynetnews ／ WIRED
Looking for holes in software used to take experts a lot of time. While people searched by hand, only so many were found. What changes when AI joins the search?
An AI helped find holes in a language that runs a large share of the web. Its developer says the rules for reporting and fixing such holes should change too. The concern seems to be that repairs cannot keep pace with discovery.
| Who | What was reported |
|---|---|
| Sam Altman (OpenAI CEO) | Warned AI is exposing flaws at alarming speed: "shocked" |
| Mistral | New Le Chonk equals or beats Chinese rivals in coding, cybersecurity |
The developer's chief executive was openly surprised by the speed. Another company claims the same skill now puts it level with rivals abroad. Finding holes is becoming a selling point in the model race. If so, defenders are not the only ones with this power. The same tools can reach attackers.
Drug companies run many patient websites and internal systems. If attackers hunt with AI, the days between discovery and repair become the size of the risk. Look for old sites left unattended, and review them more often. If a vendor runs a site for you, confirm the agreed repair time in the contract.
While defense takes more effort, money also moved around the systems that drug companies use every day.
04Money for life sciences platforms
The business systems drug companies use every day can change owners. On this day, two money moves around such systems came together. It is backstage news, yet it reaches daily work directly. What matters is not the amount but whose hands it lands in.
| Deal | Type | Amount |
|---|---|---|
| Dassault Systèmes → ArisGlobal | Acquisition completed; expands life sciences AI platform | Not in headlines |
| Allox (Catalonia) | Raised funds to advance its AI drug discovery platform | €9.4M |
Allox was the only deal that day with a disclosed amount
A large software group closed a purchase to grow its AI offering for drug makers. I read it as buying a system already in use, along with its customers. A young firm using AI to find new medicines also raised development money. So the money splits two ways: toward finding new drugs, and toward running daily operations. Yet only the small deal shows a price.
When the owner of your system changes, contract terms and future plans may change too. Where your data is stored, and how it connects to other systems, may shift. News of a purchase is a signal to reread your own contract. If you keep safety or quality records there, check that their handling stays the same. Ask early about migration dates and price changes.
Just as with business systems, whether we can trust what we see on video now needs defending.
05Finding edited segments in videos
Source AV-FABLE: Audio-Visual Manipulation Forensics in Long-Form Videos
When people hear about fake videos, they picture clips made from scratch. The harder problem is a short lie mixed into genuine footage. Because the scenes around it are real, viewers can easily miss the lie in the middle.
If most of a long clip is genuine, judging it as a whole lets the edit slip through. So researchers treated sound and picture separately. Their test asks when, and in which one, something was changed. As the tricks got finer, the checks got finer too.
Asking in words
AV-LLMs, without extra training, were prompted to say if a video was edited. Detection was near chance, and localization also failed.
Reading internal features
NFSR reads frozen internal features of the same AV-LLM. It gave the best audio and video localization results.
A pre-peer-review report
An arXiv preprint, not yet peer reviewed; abstract only. The benchmark, code and model weights are to be released.
The results were clear. Simply asking an AI did no better than guessing. But reading the inner signals of that same AI pinpointed the edits best. Still, this work has not yet been checked by other researchers.
Pharma also relies on videos, such as doctor lectures and internal briefings. If one remark were swapped, false claims about a drug could spread in the speaker's own voice. Keep the original files of what you publish. If someone suspects a swap, you can compare and correct it quickly.
Just as with video, the reports coming out of pharma work deserve the effort of checking them yourself.
06On the pharma front line
Source https://ai-hikari-ai.com/ ／ BigGo ファイナンス ／ Hunton Andrews Kurth LLP
Now we move closer to pharma work. Stories about lighter workloads are attractive. But the better the news, the more you need to ask where the numbers came from.
Regulatory document review
Anthropic reportedly published results with 6 pharma companies: review time cut by 89%, and human oversights also caught.
Sales visits into coaching
An Inizio executive said 27 top global pharma companies are piloting STEM AI. This is not the number in full rollout.
State law on medical chatbots
A California law extended CMIA protection to health care chatbots. It also set new rules for AI use in clinical care.
The claim of much faster reading of regulatory papers comes from the developer itself. We do not know which firms took part or how it was measured. The tool that turns sales calls into coaching is still only being tried by many large firms. Trying a tool and keeping it are very different things. And in one American state, confidentiality rules now cover AI that answers health questions.
Big numbers tempt us to expect the same gain at home. But a number without a known method cannot be compared with anything. Avoid copying a vendor's claim straight into internal slides.
Decision makers should run a small trial on their own files and measure it themselves. Record whether the AI caught missed errors, side by side with human checks. If you plan a tool that answers patients, first check how far confidentiality rules reach. Only numbers you measured belong in internal decisions.
That same habit of checking should also apply to where an AI itself comes from.
07Distillation has shifted targets
Source Reuters
Last, a story about where AI learns from. The answers a strong AI gives can become training material for other AIs. Seeing that flow changes how its strength looks.
| Item | What was reported |
|---|---|
| Said to have used it | DeepSeek, Moonshot AI |
| Model said to be used | OpenAI's GPT-6 Astra |
| Model said to be the earlier target | Anthropic's Claude |
| How it was reported | Reuters carried The Information's report |
Two developers in one country reportedly trained their own AI by asking another firm's newest model huge numbers of questions. An earlier report named a different company's AI as the target. The target seems to move to whichever AI is strongest at the time. This is still a press report, not a confirmed fact. If true, gaps in strength close quickly.
Gather enough answers, and you can copy an expensive skill cheaply. That is why developers try to guard how their models answer. Tighter guards may also mean stricter terms for users.
For drug companies, where an AI comes from becomes part of explaining quality. An AI of unclear origin in your workflow may leave you unable to explain it later. When you feed company knowledge into an AI, assume it could train someone else's model. Confirm in the contract that your inputs will not be used for training.
Strength alone is not enough. Asking where that power came from leads to the final answer to today's question.
Can Google win back revenue in the 10 countries where it expanded AI Mode ads? Will there be updates on fixing the PHP flaws?
Open the full transcript
Intro
News that AI has grown smarter is no longer rare. Today's stories ask what comes after that. Who earns money from that power, and who checks it?One company posted strong results and still saw its shares sold off. The ability to find holes in software is reaching attackers as well as defenders. Reports of saved effort in pharma work keep coming. But we cannot yet tell whether those figures would hold inside your own company. People have also started asking what an AI learned from. Today I want to look past raw strength. I want to look at its price and its evidence.
CH 01 Alphabet shares plunge
We begin with one of the world's most profitable companies. Its total sales grew strongly, yet investors sold the stock. They seemed to care less about overall growth than about how its core business earns, and where that is heading.The real question was how well ads can pay on a screen where AI writes the answer. A user who is satisfied by the answer may click fewer ads. Usage can grow while each visit earns less. I read the sell-off as fear of that outlook.Classic search listed where answers could be found and sent people elsewhere. Ads sat along that route. When AI writes the answer itself, the route disappears. So does the old place to put an ad.The value that vanished was enormous. Worry about a changing revenue model outweighed the strong figures.This matters to pharma teams too. When doctors or patients look for information, more searches will end with the AI's answer. The idea that people will reach your own website is weakening. Now is the time to rethink how your information travels. So is advertising the only way to earn from AI?
Another company has chosen a different path.
CH 02 Two ways to make money from AI
The same AI can cost users very differently, depending on who pays. Drug companies pay on both paths. Knowing which path a vendor takes is where price talks begin.One side places ads beside the AI's answers and is taking them abroad. The other charges for each piece of work done with its AI. It already earns large sums from a deal with a major model developer. The first earns from advertisers. The second earns directly from businesses. Neither is right or wrong. The difference is who pays.When you pay per use, the bill rises as the AI works harder. The longer the tasks it takes on, the more it costs. There are reports of plans to raise prices in line with the workload.If a pharma unit scales up AI, the first quote is not enough. Estimate the cost a year later, with more users and more tasks. Success itself can push you over budget. Set usage caps for each department, and review usage every month. Beyond pricing, the power of AI itself is starting to change how systems are defended.
CH 03 Vulnerabilities found by AI
Looking for holes in software used to take experts a lot of time. While people searched by hand, only so many were found. What changes when AI joins the search?An AI helped find holes in a language that runs a large share of the web. Its developer says the rules for reporting and fixing such holes should change too. The concern seems to be that repairs cannot keep pace with discovery.The developer's chief executive was openly surprised by the speed. Another company claims the same skill now puts it level with rivals abroad. Finding holes is becoming a selling point in the model race. If so, defenders are not the only ones with this power. The same tools can reach attackers.Drug companies run many patient websites and internal systems. If attackers hunt with AI, the days between discovery and repair become the size of the risk. Look for old sites left unattended, and review them more often. If a vendor runs a site for you, confirm the agreed repair time in the contract. While defense takes more effort, money also moved around the systems that drug companies use every day.
CH 04 Money for life sciences platforms
The business systems drug companies use every day can change owners. On this day, two money moves around such systems came together. It is backstage news, yet it reaches daily work directly. What matters is not the amount but whose hands it lands in.A large software group closed a purchase to grow its AI offering for drug makers. I read it as buying a system already in use, along with its customers. A young firm using AI to find new medicines also raised development money. So the money splits two ways: toward finding new drugs, and toward running daily operations. Yet only the small deal shows a price.When the owner of your system changes, contract terms and future plans may change too. Where your data is stored, and how it connects to other systems, may shift. News of a purchase is a signal to reread your own contract. If you keep safety or quality records there, check that their handling stays the same. Ask early about migration dates and price changes. Just as with business systems, whether we can trust what we see on video now needs defending.
CH 05 Finding edited segments in videos
When people hear about fake videos, they picture clips made from scratch. The harder problem is a short lie mixed into genuine footage. Because the scenes around it are real, viewers can easily miss the lie in the middle.If most of a long clip is genuine, judging it as a whole lets the edit slip through. So researchers treated sound and picture separately. Their test asks when, and in which one, something was changed. As the tricks got finer, the checks got finer too.The results were clear. Simply asking an AI did no better than guessing. But reading the inner signals of that same AI pinpointed the edits best. Still, this work has not yet been checked by other researchers.Pharma also relies on videos, such as doctor lectures and internal briefings. If one remark were swapped, false claims about a drug could spread in the speaker's own voice. Keep the original files of what you publish. If someone suspects a swap, you can compare and correct it quickly. Just as with video, the reports coming out of pharma work deserve the effort of checking them yourself.
CH 06 On the pharma front line
Now we move closer to pharma work. Stories about lighter workloads are attractive. But the better the news, the more you need to ask where the numbers came from.The claim of much faster reading of regulatory papers comes from the developer itself. We do not know which firms took part or how it was measured. The tool that turns sales calls into coaching is still only being tried by many large firms. Trying a tool and keeping it are very different things. And in one American state, confidentiality rules now cover AI that answers health questions.Big numbers tempt us to expect the same gain at home. But a number without a known method cannot be compared with anything. Avoid copying a vendor's claim straight into internal slides.Decision makers should run a small trial on their own files and measure it themselves. Record whether the AI caught missed errors, side by side with human checks. If you plan a tool that answers patients, first check how far confidentiality rules reach. Only numbers you measured belong in internal decisions. That same habit of checking should also apply to where an AI itself comes from.
CH 07 Distillation has shifted targets
Last, a story about where AI learns from. The answers a strong AI gives can become training material for other AIs. Seeing that flow changes how its strength looks.Two developers in one country reportedly trained their own AI by asking another firm's newest model huge numbers of questions. An earlier report named a different company's AI as the target. The target seems to move to whichever AI is strongest at the time. This is still a press report, not a confirmed fact. If true, gaps in strength close quickly.Gather enough answers, and you can copy an expensive skill cheaply. That is why developers try to guard how their models answer. Tighter guards may also mean stricter terms for users.For drug companies, where an AI comes from becomes part of explaining quality. An AI of unclear origin in your workflow may leave you unable to explain it later. When you feed company knowledge into an AI, assume it could train someone else's model. Confirm in the contract that your inputs will not be used for training. Strength alone is not enough. Asking where that power came from leads to the final answer to today's question.
Wrap-up
Today's stories pointed the same way. The stronger AI gets, the more users must check its price and its evidence themselves. How fees grow, how fast defense moves, how reports were measured, what an AI learned from. You can ask all of this before you sign. Next, we will watch whether ads beside AI answers really pay, and whether repairs to those holes move forward.
- CH 01WBUR「Alphabet stock tanks on Google's disappointing AI-search revenue」 wbur.org
- CH 01The Times「$500bn wiped off value of Google owner Alphabet after missing targets」 thetimes.com
- CH 01Tech Insider「Google Search Hits $63.4B as AI Mode Pays Less [2026]」 tech-insider.org
- CH 01The Tribune「Alphabet revenue jumps 25% to USD 127.6 billion as Google Cloud growth doubles」 tribuneindia.com
- CH 02blog.google「Ads in AI Mode expand to new markets.」 blog.google
- CH 02CRN「Microsoft Q1 Earnings: CEO Nadella Touts AI Consumption, Usage-Based Billing Growth」 crn.com
- CH 02TechCrunch「Microsoft recorded $10 billion in revenue from OpenAI」 techcrunch.com
- CH 02Benzinga「Microsoft Wants to Charge More as AI Agents Do More Work, While Google Sees a New Advertising Gold Mine」 benzinga.com
- CH 03Mashable「OpenAI says GPT-5.6-Sol helped find the PHP flaws now shaking the internet」 mashable.com
- CH 03Ynetnews「'I'm shocked': OpenAI CEO Sam Altman warns of AI systems uncovering alarming security vulnerabilities」 ynetnews.com
- CH 03WIRED「Mistral Says Its AI Model ‘Le Chonk’ Has Leapfrogged China in the Open-Weight Race」 wired.com
- CH 04HPCwire「Dassault Systèmes Completes ArisGlobal Acquisition to Expand Life Sciences AI Platform」 hpcwire.com
- CH 04Dealroom「Catalan biotech Allox raises €9.4M to advance AI drug-discovery platform」 app.dealroom.co
- CH 05AV-FABLE: Audio-Visual Manipulation Forensics in Long-Form Videos（「この種の改ざんは、動画全体が本物か偽物かを判定しても見つからない。」）
- CH 06https://ai-hikari-ai.com/「Anthropicが製薬6社との協業成果を公表｜規制文書レビューの時間を89%削減・人の見落としも検出——久留米・福岡の中小企業向けAI活用ガイド」 ai-hikari-ai.com
- CH 06BigGo ファイナンス「Inizioのニック・サインダーコム氏、世界大手製薬27社が営業訪問をコーチングに変えるSTEM AIを試験導入中と発言」 finance.biggo.jp
- CH 06Hunton Andrews Kurth LLP「California Law Extends CMIA Protections to Health Care Chatbots and Enacts New Guardrails on AI Use in Clinical Care」 hunton.com
- CH 07Reuters「DeepSeek, Moonshot used OpenAI's GPT-6 Astra to train latest models, The Information reports」 reuters.com
Articles used
- AI Daily News 2026-10-30
- AI Daily News 2026-10-29 Evening
- AI and the Pharma Industry 2026-10-30
- AI Daily News: Article Summaries (Oct 30 Morning)
- AI Daily News: Article Summaries (Oct 29 Evening)
- AV-FABLE: detecting audio and video edits in long-form videos
Today's related reports
- AI Daily News October 30, 2026
- AI & Economy News October 30, 2026
- AI & Finance News October 30, 2026
The narration is synthetic speech. The content draws only on this site's articles from the same day. Each edition passes seven plain-language checks before publication. That means known obstacles to comprehension are held below threshold — it is not a guarantee of comprehension.