AI Highlights — the whole picture — 2026-11-11 (Wed) Daily News

Qantas reportedly signed an AI contract with Accenture and cut approximately 850 jobs. Around the same time, AI stocks plummeted in the wake of the election, and banks and investors became wary of AI-related debt. In drug discovery, funding and collaboration announcements continue to precede results. Both personnel and funds are being mobilized before the effects of AI can be confirmed in numbers. Explanations such as ``It's because of AI'' or ``Thanks to AI'' are likely to be prepared before reading. As I read this half-day article, I wanted to find out what the causes of each report were and whether those causes were verified. The fact that a connection has been reported and the fact that cause and effect have been confirmed are treated as different things.

Is the reason for personnel reductions AI or the economy?

Australian airline Qantas has reportedly cut around 850 jobs after signing an AI deal with Accenture (Quartz India). Because the company name, number of people, and contract are listed, it's tempting to read it as an example of AI stealing jobs. But this report goes so far as to link cuts to AI contracts. It is unclear from the information in the title how much of the work the contract will replace or how many of the 850 employees were directly affected by AI. There is a difference between there being a connection and confirming that AI is the cause.

There is another view. Analysts say the economy, not AI, is the main driver of tech layoffs (TradingView). Even though the word "reduction" is the same, the causes are different between what happened alongside the AI contract and what happened as a result of the economic downturn. When these two reports are placed side by side, there are points that readers can confirm. The three points to consider are whether the reductions were announced before or after the introduction of AI, whether the companies themselves were citing AI as the reason, and whether there was any evidence of economic deterioration during the same period. It can be said that explanations based on AI can only be verified after such confirmation.

There are also movements surrounding human resources. The Singapore Economic Development Board (EDB) shares expert views on generative AI and the future of work. Universities are reportedly incorporating AI into classes without hiring new students as companies seek AI skills (Higher Ed Dive). The concrete fact is that the existing system is being used without increasing the number of teaching staff. These moves are in line with the promise that AI will change the nature of work, but they do not mean that the promise is backed up by employment statistics. It is appropriate to accept the opinions of experts as nothing more than opinions.

What we can say from the material so far is that both reductions and human resource development are underway before the effects of AI are confirmed. In terms of employment, explanations for the cause are divided. So, is the same precedent occurring on the side of investing money, for example, in investments in drug discovery?

Plummeting market and the politics of regulation and conflict

AI stocks plunged in the wake of the election, and U.S. Treasuries gained momentum (Bloomberg.com). At the same time, banks and investors are taking a cautious stance on AI-related debt. Reuters reports this as a move in response to the market plunge. This means that all three markets - stocks, government bonds, and debt - have become increasingly focused on AI-related assets at the same time. The decline in stock prices indicates a reassessment of valuations, and the cautious attitude toward debt indicates that those providing funds have begun to look more closely at the prospects for recovery. However, it is not possible to ascertain the reason for selling or taking a cautious stance just by looking at the title of each article.

Movements also overlapped on the political side. The New York Times reported that Democrats have proposed tougher AI regulations, and the question is what that will mean for the economy. NBC News reports that Nvidia CEO Huang has been accused of Chinese espionage, which has become a key point in the U.S.-China AI conflict. The allegations are at the stage of being reported and should not be treated as a conclusion. Still, the fact remains that regulatory debates and US-China tensions are on the same day as news about stock prices and debt. For companies working on AI, the strength of regulations and relations with China are directly related to the conditions for proceeding with business.

What readers should reconsider is the assumption that AI investment is driven solely by technological results. If the attitudes of funding banks and investors, the positions of the ruling and opposition parties on regulations, and the relationship between the United States and China change, the profitability of investment could change before the effects of AI can be confirmed in numbers. The personnel cuts we saw in the previous section were also determined under these financial and political conditions. Before the effectiveness of AI has been verified, funders and political parties are taking the lead. In the next section, we will look at the order in which investments in drug discovery proceed under these conditions.

Pharmaceuticals and actual implementation in the field

In drug discovery, funding and collaboration announcements continue. Sanyou Bio will expand its AI-powered drug discovery platform (Pulse 2.0) with approximately $15 million (RMB 100 million) Series C funding. 3billion has signed a collaboration agreement with DeepLife for AI drug discovery (3billion). FRONTEO and Tagsix have begun joint development, the second round of AI drug discovery (news media VOIX). What we can confirm is the business movement of funding, contracts, and the start of development. From these presentations, it is not possible to tell what has been achieved as a result of drug discovery. What readers should distinguish between is whether the announcement is referring to funding or a contract, or to the results of a drug.

Major pharmaceutical companies are taking concrete steps to improve their business infrastructure. AstraZeneca has expanded its partnership with SAP to modernize its pharmaceutical business and leverage AI (Manufacturing Chemist). SAP announced that this expanded collaboration will promote the introduction of RISE with SAP and Business AI (SAP News Center). PSC Biotech partners with Clever Culture Systems to provide AI-enabled microbial testing automation to pharmaceutical manufacturers (EIN News). Projects include integration into existing business areas such as core systems and inspection processes. However, these announcements describe the promotion and provision of implementation. Numerical values showing the effects after introduction are not included in the materials at hand.

Cases at the same stage can be seen outside of pharmaceuticals as well. DALTEX, the world's leading potato exporter, has implemented ORTH and AI on a large scale (GlobeNewswire). On the other hand, there is also a report that AI analysis of electrocardiograms from smartwatches may be able to detect a decrease in heart contractility (www.heart.org). This is only at the stage where the expression is ``detected,'' and it needs to be read separately from cases that have already been introduced.

To summarize, the announcement of implementation will go first, and the verification of effectiveness will come later, or has not yet come. In the next section, I would like to confirm whether this order also applies to employment reductions.

Research to measure the reliability of memory and learning data

Agents who complete long tasks over many days have a memory system that allows them to remember what they did last time. In a pre-peer-reviewed preprint published on arXiv, MemoryDojo proposed a benchmark to measure how malicious instructions entered into memory perform in subsequent sessions (on their site). Instructions read one day are stored and retrieved and executed another day. This study compares how memory poisoning works differently depending on the storage method. The authors argue that damage occurs differently depending on the method, and that it can be measured. On the other hand, in the abstract range, it is not known which models and storage methods were used and how many attacks were tried. Even if numerical values are presented, they cannot be directly applied to your company's operations.

The remaining two questions question the correctness of the training data. When retraining a language model using data that does not have a correct answer, there is a way to have the model answer many times and use the majority answer as a pseudo label.テスト時強化学習（TTRL）である。 The arXiv preprint shows a method to divide the majority vote error into two and suppress only the side that can be handled without bringing in the correct answer from outside (own site). Another, AbCD, proposed a method to have multiple models discuss and turn the records into ``favorite data'' without labeling the correct answer (on their own site). Normally, to later train a model that is good at inference, you need a system to determine whether the answer is correct, and a learned reward model. AbCD is designed for situations where this is not available. However, both papers have not yet been peer-reviewed, and it is not possible to ascertain how well they will pass for each subject based on the abstract alone.

What these three cases have in common is what they believe, learn, and put into practice. Memory believes past input, majority vote believes the most popular answer, and record of discussion believes the result of discussion. In any case, if what you believe is wrong, that mistake will remain in your next action or learning. Safety evaluation can also be read as requiring measuring not only the model's answer, but also the memory and label that led to that answer.

On the enterprise side, BlackLine announced AI plans in line with its partnership with SAP (TradingView). All that can be seen from the headline is that a plan based on the partnership has been announced. It is not clear from this headline how the reliability of memory and learning data is verified. The plan will be announced first, and its effectiveness and safety will be verified later. This order has the same form as the employment and investment stories we saw in the previous sections. The decision to implement AI may be made before the means of verification are in place.

In applications that relate to people's concerns, the risks are considered before the benefits. When asked whether AI can replace therapy, experts at the University of Alabama at Birmingham (The University of Alabama at Birmingham) explained both the risks and benefits. Regarding the movement to include AI in apps for caregivers, it has been pointed out that this alone may not be enough (WIRED). In the area of selling girlfriend roles, an article about ``Girlfriend Fake'' has been compiled into a single article about AI apps, pranks, rentals, and how to identify fraud (Shirasu Izumi). This means that the supporting function and the deceptive business are side by side under the same name of ``snuggling''.

Attention is also being paid to the demand side. A WSJ editorial asked why people would want AI friends (WSJ). Technology to read emotions is also included in the game, and it was reported that AI that changes the in-game weather according to the player's facial expressions will enhance the immersion (Bioengineer.org). Technology that handles personal information, such as facial expressions, is expanding as an entertainment function. Additionally, Microsoft AI's CEO reportedly said that he would consider AI rights if there was evidence of consciousness (Futurism). This statement is not an admission that he is conscious, but can be read as a conditional statement if there is evidence.

The question of responsibility is also directed towards the developer side. There has been commentary (The Washington Post) that AI companies are trying to give chatbots a conscience, and giving machines a basis for decision-making has become inseparable from the question of whose values are being included. In terms of prejudice, when asked whether AI is racist, a lecturer explained how algorithms can strengthen society's prejudices (The Brussels Times). The more people-friendly functions become available, the more people are likely to receive biased responses or inappropriate advice.

What these articles have in common is that, before examining their effectiveness, they raise questions about the spread of use and responsibility. What readers should look at is not only what AI can do, but also who should take on what responsibility. Drawing that line will also serve as the basis for the next discussion on systems and costs.

What these half-day articles had in common was that the movement came first, and the confirmation of the effects was left behind. Layoffs were reported alongside both AI contracts and the economy, and the reason for the market crash and cautious stance on debt could not be ascertained. Announcements about drug discovery only showed funding and contracts, but did not show results. Research on memory and learning data was at the stage where a means of measuring it had been proposed. In each case, what the news reports are reliably reporting is the event and its sequence, not a determination that AI was the cause. The explanation that "it's because of AI" needs to be reconsidered based on two points: what the cause is, and whether it has been verified. It is appropriate to treat explanations before numbers are provided as provisional explanations.

Q1: Of the things that you say have changed in your work, how many of them have been verified by looking at numbers before and after the introduction? Q2: When receiving approval or reporting on the use of AI, do you check who has conducted the verification that provides the basis for its effectiveness and safety, and to what extent? Q3: Can you differentiate and explain whether the reasons for investing in AI or making personnel decisions were the verified results, or the surrounding movements and announcements?

Finally, three questions

- How many of the things that you say have changed in your work that have changed because of the introduction of AI, have you verified them with numbers before and after the introduction? Q2: When receiving approval or reporting on the use of AI, do you check who has conducted the verification that provides the basis for its effectiveness and safety, and to what extent? Q3: Can you differentiate and explain whether the reasons for investing in AI or making personnel decisions were the verified results, or the surrounding movements and announcements? - When receiving approval or reporting on the use of AI, do you check who has conducted the verification that provides the basis for its effectiveness and safety, and to what extent? Q3: Can you differentiate and explain whether the reasons for investing in AI or making personnel decisions were the verified results, or the surrounding movements and announcements? - Can you differentiate and explain whether the reasons for investing in AI or making decisions about personnel were the verified results, or the surrounding movements and announcements?