Summary
There is an in-depth discussion in an MIT Technology Review article on the challenge that AI chatbots are making to search engines. Data shows that Google Search is more popular than ever. It has integrated AI through the AI Overviews feature, and advertisement revenue is strong. ChatGPT on the other hand offers a conversational search, which seems appealing on the surface, but which does not yield nearly the same revenue for OpenAI. In addition, chatbots are facing opposition from content producers due to the zero-click phenomenon – where AI chatbots recover data from sites, even from behind paywalls, and allow users access to this data without the user ever visiting the website.
At the start of the year, articles are looking back on 2024 and forward to 2025. One article mentions AI slop, highly publicized hallucination cases, and the proliferation of non-consensual porn as the main failings of AI in 2024. An O’Reilly report cites Claude and Llama as among the most popular AI models, and retrieval-augmented generation (RAG) and LangChain (for AI agent orchestration) as the most sought-after emerging technologies of 2024. Deloitte’s latest State of Generative AI in the Enterprise report cites data security fears and regulatory uncertainty as being among the main roadblocks to successful AI projects. Finally, a VentureBeat article predicts that agent orchestration frameworks like LangChain and LlamaIndex will be highly popular in 2025.
On the cybersecurity front, Chinese hackers are accused of breaching the U.S. Treasury office and stealing documents. Meanwhile, Ivanti announced that its flagship VPN products are currently the target of a zero-day exploit.
For Big Tech, Nvidia made a series of announcements relating to chips and services for AI. The CEO announced that Nvidia will be at the heart of a world with one billion humanoid robots, 10 million automated factories, and 1.5 billion self-driving cars and trucks. Meta is facing fierce criticism after it decided to remove human fact-checkers for its platform, paving the way for what Maria Ressa, the Nobel peace prize-winning journalist, calls “extremely dangerous times” for journalism, democracy and social media users.
Table of Contents
1. AI means the end of internet search as we’ve known it
2. The biggest AI flops of 2024
3. Why 2025 will be the year of AI orchestration
4. Chinese hack of US Treasury breached sanctions office, Washington Post says
6. Hackers are exploiting a new Ivanti VPN security bug to hack into company networks
7. Meta’s changes to policing will lead to clash with EU and UK, say experts
8. CES 2025: Nvidia unveils gaming chips, desktop computer to protect AI lead
9. Deloitte’s State of Generative AI in the Enterprise - Quarter three report
1. AI means the end of internet search as we’ve known it
This article examines the emergence of AI when searching the Internet for information, and how AI chatbots (like ChatGPT) are seen as challenging search engines. This is especially the case for Google Search – by far the most widely used search engine. AI offers the possibility of a conversational search, where information is presented in a structured format instead of a list of links, and where the user can drill down to gain more specific answers. An example query which appears more suited for AI is “I’m going to Japan for one week next month. I’ll be staying in Tokyo but would like to take some day trips. Are there any festivals happening nearby? How will the surfing be in Kamakura? Are there any good bands playing?”. Compared to a traditional web search query, the user is far less specific about the information being looked for.
Despite the appeal of AI, data shows that Google Search is more popular than ever, partly because a lot of Web searches seek to find current information while AI chatbots have a cut-off date for their training data. That said, Google has been integrating AI into its search engine since May 2024 with the appearance of AI Overviews. Built on their Gemini model, this deployment had an agitated start with the much publicized error of telling people to eat rocks and to use glue in their pizzas. Google has since deprecated the use of less reliable data sources, like Reddit. Further, it has linked the system to its Knowledge Graph – a database of trillions of facts about the world. Another improvement that AI is bringing to Google Search is the possibility of searching based on an image (Google Lens) and now a video. Finally, tools like NotebookLM will allow content to be created (e.g., videos) on the fly to explain results in links.
The big worry for websites and content publishers is the zero-click scenario – where AI chatbots recover data from sites, even from behind paywalls, and allow users access to this data without the user ever visiting the website. It should be noted that Google Search has already some zero-click features; for example, when searching for a shop, Google displays the opening hours of that shop for the user, who therefore does not need to visit the shop’s website. Nonetheless, Google relies on users visiting websites for its advertisement revenue. On the other hand, 250 million people use ChatGPT on a weekly basis, yet, OpenAI is projected to lose 14 billion USD in 2026. Another cited advantage with search engines is that the user understands why a particular link was returned in a search. It is harder to explain why a chatbot’s model chooses a particular information source in response to a prompt query.
2. The biggest AI flops of 2024
This article looks at some of the failings and underachievements of AI in 2024. The first is the proliferation of AI slop (content generated by AI, and found in all places from newsletters to Amazon advertisements). One problem with slop is the model collapse – the phenomenon that a model trained on AI generated data degrades in performance. Slop is easy to generate on a large scale and is often used to generate pictures with an emotional impact in order to encourage sharing on social networks (e.g., content related to the Israel-Palestine conflict). A second failing is that AI is impacting expectations of real events; the article cites the case of a Halloween parade in Dublin that was announced in an AI generated article (from Pakistan). Hundreds of people turned up for the event though no parade had ever been planned.
Another underachievement cited is Grok – the image creation tool from Elon Musk’s xAI. This tool does not have the guardrails of other image generators (which Musk calls “woke AI”) apart from a block on generating nude images. Another failing of 2024 was the use of Microsoft’s AI image generator to create sexually explicit deepfake images of Taylor Swift (a US singer). This case highlighted the difficulty in combatting non-consensual porn in general. A final problem cited are the errors generated by AI tools. For instance, a chatbot for New York city set up to advise citizens hallucinated and gave answers to questions about how to break the law.
3. Why 2025 will be the year of AI orchestration
Agents and agentic behavior were among the buzzwords of 2024. This is the idea of a chatbot acting on a user’s behalf, for instance, to buy a product for the user and even handling the payment. This VentureBeat article argues that in 2025, the challenge will be to develop and deploy platforms that can orchestrate several agents. LangChain is one such platform that has become very popular, though Microsoft’s Magentic and LlamaIndex could become quite competitive this year. One challenge for agent frameworks is the transfer of context across applications. In this regard, AWS’s Bedrock is cited for its ability to provide agent connections to Salesforce’s Agentforce and ServiceNow. Executives are keen that this year show a return on AI investment, and agents are seen as the shortest route to this. Nonetheless, one expert is cited as saying that user acceptance of AI tools is still the hardest challenge for uptake of the technology as many employees still prefer manual methods of working.
4. Chinese hack of US Treasury breached sanctions office, Washington Post says
Chinese hackers breached the U.S. Treasury office and are believed to have stolen documents from the Office of Foreign Assets Control and the Office of Financial Research that relate to economic sanctions on companies. The breach is believed to have been through BeyondTrust, a cybersecurity provider for the Treasury. The attack comes in the context of a tense relationship between China and the US. In particular, the US Treasury is thinking about economic sanctions against Chinese banks because of China’s aid in funding Russia and its war in Ukraine. The Treasury has described the hack as a “major incident”. For its part, China has described the hacking claims by the US as “irrational” and “without any basis”, and represents a “smear attack” on the Chinese government.
5. Technology Trends for 2025
This report from O’Reilly looks at technology trends in 2024 (January to September), based on user demand for content on O’Reilly’s Learning Platform. For language models, Claude is now the favorite model of programmers. Google’s Gemini is also popular, as is Llama because the models of this family are open, and many can be run on laptops. Over the year, demand for learning content on Machine Learning grew 9.2%, the demand for Artificial Intelligence grew by 190%, Natural Language Processing grew by 39%, Generative AI grew by 289%, and Prompt Engineering grew by 456%. The report notes that the increased demand for prompt engineering may be temporary, because as model reasoning improves, the need for clearly structured prompts or “explain it to me like I’m a ten-year old” will decrease.
The demand for GitHub Copilot increased 471% in 2024. On frameworks, two emerging skills in 2024 are retrieval-augmented generation (RAG) and LangChain. RAG interests users because it allows systems to gain access to up-to-date, company-specific documents, and also because it reduces the risk of AI hallucination. LangChain is demanded because it is a framework for building and orchestrating agent applications. On that point, the demand for Enterprise Architecture is up 17% and Distributed Systems is up 35%. On programming, O’Reilly estimates that programmers actually spend roughly 20% of their time writing code. The remaining time is spent on testing, design and documenting. The demand for Software Architecture rose by 5.5% in 2024. This may be because junior programmers are afraid of job layoffs, and want to rise to a more senior position to avoid this. The report notes that the demand for Micro-services declined by 24% and Domain-Driven Design by 22%. It postulates that many organizations do not have the competence to implement micro-services, so the demand is actually failing.
In cybersecurity, Security Governance was the most requested topic, even ahead of Network Security. This is because companies seem to be more aware of the reputation damage of cyberattacks. The demand for Zero-Trust also rose this year.
6. Hackers are exploiting a new Ivanti VPN security bug to hack into company networks
Ivanti, the US software company, has announced that there is a zero-day vulnerability in its Connect Secure, Policy Secure, and ZTA Gateways products. The Connect Secure remote-access VPN solution is, according to Ivanti, “the most widely adopted SSL VPN by organizations of every size, across every major industry”. The vulnerability, tracked as CVE-2025-0282, allows an adversary to deploy malicious code on the Ivanti products. Several companies are believed to have been compromised by the zero-day exploit, which is thought to have been active since mid-December 2024. A patch for all devices will not be made available until January 21st. Ivanti did not say who the hackers might be, but the security firm Mandiant suspects a China-linked cyber-espionage group.
7. Meta’s changes to policing will lead to clash with EU and UK, say experts
Meta announced that it is abandoning human fact-checkers for its social networks Facebook, Instagram and Threads, in the US at least. The company says that it will allow the social media users to auto-regulate content using community notes, a similar system to what is used by the X platform. Content about suicide, self-injury and eating disorders will still be considered “high-severity violations”, and “automated systems will scan for that [...] content”. On the other hand, the company states that they “allow allegations of mental illness or abnormality when based on gender or sexual orientation”.
Meta’s decision is facing fierce criticism. Maria Ressa, the Nobel peace prize-winning American-Filipino journalist, said the decision is a precursor to “extremely dangerous times” for journalism, democracy and social media users. In the UK, a spokesperson for the parliamentary committee which is investigating the impact of on-line dissemination of disinformation on the riots that occurred last Summer called the decision “quite frightening”, adding that “people have a right to be protected from the harmful effects of misinformation... The fact that Zuckerberg said he’s following the example of X must raise concerns when we compare how X is a platform for misinformation to a greater extent than Facebook has been.” Meta’s decision is also seen as a reaction to the election of Donald Trump to the US presidency. There is speculation that Trump will repeal the Online Safety Act.
8. CES 2025: Nvidia unveils gaming chips, desktop computer to protect AI lead
The Nvidia CEO, Jensen Huang, made a series of announcements relating to chips and services for AI that the company is working on. Nvidia is updating its GeForce GPUs with the Blackwell GPU design used by its AI accelerators. Among other things, this will lead to more realistic gamer experience as images are not only constructed by calculating the shade of each individual pixel, but also by having AI techniques predict in advance what the next frames should look like. The RTX 5090 and RTX 5070 models will be available early in 2025, costing 1999 USD and 549 USD respectively. Nvidia is cooperating with MediaTek on Project Digits – a Linux based desktop PC with a single Grace Blackwell Superchip that allows powerful models to be run locally by developers. The PC costs 3’000 USD.
The CEO announced that Nvidia will be at the heart of a world with one billion humanoid robots, 10 million automated factories, and 1.5 billion self-driving cars and trucks. Nvidia is cooperating with Toyota and Uber for autonomous driving AI products that use Nvidia’s Drive chips and software. The article notes that data center revenue for Nvidia will be around 100 billion USD in 2025, exceeding its gaming revenue. For instance, Microsoft’s 80 billion USD data center plan will mostly use Nvidia chips.
9. Deloitte’s State of Generative AI in the Enterprise - Quarter three report
Deloitte has published its quarterly review on the state of Generative AI in the enterprise. There is increasing pressure in companies to achieve a return on investment on AI, though an acceptance that change management is not just about technology, but also about people and processes. This quarterly report focuses on data, governance, and compliance risks. The authors surveyed 2’770 director- to C-suite-level executives across six industries and 14 countries between May and June 2024. Among the key findings are that for 42% of executives, improved efficiency, productivity and cost reduction are the main drivers of AI projects. Other drivers are emerging such as increased innovation, improved products and services, and enhanced customer relationships. Nonetheless, the effectiveness of AI projects is still questioned, with 70% of organizations having moved 30% or fewer of their Generative AI experiments into production.
The main AI roadblocks cited in the survey are data preparedness (for 55% of organizations) where there are fears of leaks of personal or corporate data. Regulatory uncertainty, notably in the context of the recent EU’s AI Act is also cited as a roadblock. The lack of satisfactory KPIs for AI project success is also cited as a problem, with 41% of organizations struggling to define and measure the exact impacts of their GenAI efforts. Despite regulatory uncertainty, 78% of respondents agreed that more governmental regulation of AI was needed. Finally, 75% of organizations have increased their technology investments.