some recent GenAI news worth catching up on

Some news worth catching up on:

Antropic launched Claude Sonnet 3.5
- Competes with Open AI’s GPT-4o on price and capability.
- Important because it shows that OpenAI does not have a monopoly on high quality models.
- Therefore we need to keep evaluating multiple models for our needs.
Copyright risk for use of LLMs for software has not yet emerged.
- Current class action lawsuit has had most of its claims dismissed in an ongoing dispute.
- Important because this is a big area of use of these models.
Apple has announced the integration of LLMs into their operating system from 2025 onwards.
- Important as many LLM features are about to become expected by users.,
- Important as it shows how useful offline and local models can be combined with user data.
- Therefore we need to continue evaluating how these technologies may affect our offering.

Three recent reports to consider:

Oxford University Press (OUP)
Elsevier survey
- Results of the OUP and Elsevier reports are unsurprising, but confirm ongoing trends that we are seeing.
  - More than half of researchers surveyed had tried these models in some capacity.
  - Balance between hope for productivity gains, vs fear of large orgs, and misinformation.
  - Indian and Chinese researchers have a higher expectation of using these tools regularly within the next two years.
https://www.bondcap.com/report/aiu-e/#view/1
Mary Meeker’s report on AI and Education report supporting slides
- This report is notable because her annual internet trend reports were hugely influential in silicon valley, and she has not written a report since 2019.
- The report is less interesting than the fact that she has written one.
- She indicates that AI’s impact on higher education will be significant - which is expected, but as many of our products and services are situated within the academic context, any significant change to how that sector operates can be expected to have a follow on effect for our markets.
- Disruption drivers from the report:
  - Methods of learning have changed rapidly - GPT will accelerate this.
    - Tools that provide real-time feedback on engagement and skill development will continue to improve, enhancing the evolution of pattern recognition. AI will increasingly take over many rudimentary tasks, and the ways teachers teach, and their students learn will evolve.
  - The overnight success of LLMs has been decades in the making
    - Large Language Models (LLMs) benefit from the roughly 5 billion global Internet users who have driven an acceleration in the volume of accessible data. International Data Corporation (IDC) estimates that 163 Zettabytes of data will be created and replicated in 2024, up 80x since 2010.
  - Opportunities for individual revenue generation exist that didn’t exist before
    - It’s important for universities to understand these media and monetization shifts both for teenagers and twenty-somethings, and for the best teachers whose skills can be amplified (and monetized) off campus. What happens in media and sports may well roll into other disciplines. For better / worse, connectivity, transparency, fame, money, and short-termism can be tough competitors.
  - Debt burden of education is rising
  - Concludes:
    - “Ultimately, bringing AI to learning and teaching requires what Sal Khan calls “educated bravery.” While technologies developing in real-time are always unpredictable, their thoughtful use may well prove exponentially beneficial to students and teachers alike. We should not be paranoid and restrictive about utilising these technologies, but thoughtfully curious.”