AI Musings for May

Image generated by GPT-4o in based on a prompt from Keir Bowden

The relentless pace of change

If keeping up with the latest trends in the AI space is exhausting for us tech types, spare a thought for those outside the industry and the confusion they must be facing.

Barely a week goes by that a new best ever model appears, and is almost immediately dethroned by the next one - anyone remember Deepseek? And, of course, this isn’t helped by the fact that the latest and greatest one is often a new version of something that was mocked terribly in an earlier incarnations - Google Gemini, for example.

Take, for instance, prompt engineering. Go back a couple of years and there was a consensus that Prompt Engineer was the career of the future - the hottest job in tech where vast compensation awaited. Now advertising for a prompt engineer, outside of an LLM vendor, would be as unlikely as advertising a job driving Excel. It’s turned out to be a skill rather than a role, and everybody is expected to be able to pick it up. Plus the prompts are moving towards the internal actions of Agentic AI as opposed to something the end user has to spend hours crafting.

Unless you are treating it as a full time job, I don’t think it’s possible to stay on top of everything through your own efforts. I find it much easier to stand on the shoulders of giants and consume the curated output of those who are willing to put something approaching that level of effort in on my behalf.

My current favourites are:

• This Day in AI Podcast - this has quite a snarky and cynical tone about it, which the presenters also apply to themselves and their show. Don’t let that put you off though, as it’s full of real insight and valuable information

• TLDR AI Newsletter - this is great if you are time poor as it is extremely concise and sent out every day

• The Deep View Newsletter - another somewhat concise daily newsletter, but this one has a bit more in the way of analysis and looking to tie the various innovations together. It does tend more to the technical too

As if to prove the point, as I was preparing to hit the publish button the news broke that Instacart CEO Fidji Simo was leaving to become CEO of Applications at OpenAI, just after OpenAI gave up on efforts to hand over control to its for-profit arm. Is Sam Altman losing interest if he can’t exert the level of control he wants? Nobody (apart from Sam Altman, of course) knows, but there will be numerous column inches devoted to it and maybe throw some wobbles into pending decisions.

Agents are people too!

Switching gears slightly, earlier this year Marc Benioff predicted this is the last generation of CEOs to manage an all-human workforce. Leaving aside whether I agree that is the case, and glossing over OpenAI hiring service cloud devs rather than doing everything with their and Salesforce’s AI, I’m struck by the continuing efforts across the industry to anthropomorphise AI.

I think this is happening for a few reasons:

• Proposing AI as a helpful co-worker lowers the threat level. You can stop worrying about the mundane repetitive tasks and focus on the work that has a greater impact. This narrative is far more comforting than a Terminator needing “Your clothes, your boots, your motorcycle and your job.” And yet it’s at odds with what is happening at organisations like Duolingo, which is going AI first and replacing contractors with AI, even though it appears to be resulting in an inferior product. This also appears to be an example of AI taking the creative rather than the mundane work.

• Positioning AI as just another member of the workforce makes it easier for non-technical people to understand. While not everyone will be au-fait with machine learning, they’ll certainly grasp the concept of a colleague.

• Giving AI it’s own “persona” makes it easier to shift blame - these things are autonomous and make their own decisions, therefore it’s not our fault if something goes wrong (cynical reason #1)

• Treating AI agents as equivalent to people gets us used to the idea of spending similar amounts of money on running a tool as paying a salary (cynical reason #2)

As you can see, I view this as a combination of psychological safety, marketing and money-making - I doubt this is the first time that these three bedfellows have been sent into action together, and I’m sure it won’t be the last.