AI took my job | Issue 9

Welcome to Issue 9 of unfolding:ai, this week we are going to start to explore the impact of AI on the future of work.

A few well considered reports and studies have been released in the last week, I have pulled out the key insight from those, and added our thoughts on how they will affect us all.

You may remember that a couple of months ago, there was a large amount of press coverage over the decline in usage of chatGPT.

‘It had peaked’ vs ‘Students have gone home for the summer’

Well the data is in, Summer break is over. Minecraft usage is back down, chatGPT usage back up and above before.

I hope you enjoy this week’s edition, and don’t be shy to try out our new recommend to a friend scheme.

Best regards,

Paul, Co-founder (and newsletter editor)

AI took my job… to the next level

It’s been estimated that as much as 45% of all activity on the fiverr platform will be impacted by AI. Initial pessimistic thoughts were that this would just be a straight replacement and automation, basically removing work for its members as clients used AI themselves.

However there has been dramatic improvements in work quality, flow and responsiveness reported by the workers on the platform that are using AI. Clients seem to be delighted with the uplift. Perhaps a reflection on the customer base being value and rapid response orientated.

Typical data is suggesting, 15% increase in speed of work, and a significant 40%, ability to perform more complex tasks / activities satisfactorily. The outlook is looking strong at the moment for these marketplaces with respect to AI empowerment.

In short, AI is driving growth.

This uplift in productivity and output quality, is reflected in the findings of a recent study using Boston Consulting Group Consultants as ‘test subjects’.

This report found that the greatest impact on the workers was the ‘lower quartile in skilled’, being able to act ‘upwards’. Up to 25% faster and 40% quality boost in their output.

More of a deep dive on this report in the premium section.

Not in my job

Recent studies and research have discovered that AI is actually strong in some of the areas that were considered likely to be not in scope of the current AI technology.

LLM’s (like chatGPT) have passed creativity and innovation tests, at about 80% of human capacity. The outputs tend to be less wide and less far reaching, but in review, versus the majority of the control suggestions from humans, they scored equal or better. The absolute best results was in using AI to augment the process.

AI’s are also good at complicated tasks, specifically any that involve large amounts of data processing. It over indexes on this versus some of the ‘lower grade’ tasks typically taken by PhD workers.

C-Suite… beaten on decision making, and speed of decision with the same amount of available data for positive outcomes. Also coupled with supporting ‘act upwards’ where lower level staff can truly be empowered and enabled, it’s time for C-Suite Strategy and Leadership (at a people level) to step upwards.

Software engineers, specific code models can now achieve greater than or equal to professional experienced engineers in all major languages. With the exception of the most complex tasks. Providing they are driven by software engineers… 90% of which are using AI in some form (according to github stats).

All roles are having to adjust as to ‘what is the normal’, generally extending activities in both directions of skill, but value upwards.

The So What?

The summary impact is, looking aligned to the initial estimates by the World Economic Forum, IBM, AMD, Harvard, Gartner etc. It’s oscillating around the same numbers in each study.

Spectrum of impact is broad on knowledge workers. It is indexing more than expected by helping everyone be effective above previous levels. 15% to 40% productivity gains.

Access and adoption of chatGPT and equivalents (at a private individual level) exceeds Enterprise adoption. It is currently in ‘shadow IT’ whilst CIO’s figure out what to do (HR directors need to step in here, before we leap to ‘banning’, and cement an underground culture for AI.)

Versus ‘going it alone’ self taught users. A typically ‘trained’ user by someone with greater experience, and mentor support will outperform in the the use of chatGPT by a further 5%-8% versus untrained users.

To move from current ‘adoption’ which is at circa 13% of workforce, to 16% (which moves out of early adoption) and then onto 19% which is the tipping point towards adoption en-masse, another million users needed in the UK. Even then there will be a critical skills shortage.

Business needs a re:skill plan, and needs to think through a generational platform change, similar to the launch of office productivity, or the internet into business practices.

Spread the news

We really appreciate that you are one of our readers, here is a personalised link for you to share! Earn some digital posters, and a 1-2-1 with the founders. More offers to be added.

dalle-3 integrated into chatGPT

Next generation text to image

This just made the deadline for the newsletter No one talks much about dalle-2 but they will about this

I bet there will be game changer posts about this today/friday

Note the integration into chat to multi mode

Note the text working

Note the consistency of characters and style

This will trigger a rush on the other players now

Video here

Full announcement here 

chatGPT, adding reason before answer

These two prompts are nearly identical. But the second prompt is much more likely to provide a good answer;

Prompt 1: [Problem/question description] State the answer and then explain your reasoning.

Prompt 2: [Problem/question description] Explain your reasoning and then state the answer.

Here’s why: An LLM generates output by repeatedly guessing the most likely next word (or token). So if you ask it to start by stating the answer, (prompt 1) it will take a start at guessing the answer and then try to justify what might be an incorrect guess.

In contrast, prompt 2 directs it to think things through before it reaches a conclusion.

This also helps to explain why prompts including ‘step by step’, or ‘let’s take a deep breath and think about this before starting’, are impactful. You are helping the model to perform tree of thought reasoning.

This leads me onto one of the most interesting prompts I have seen in a while. It allows for the dynamic creation of an interactive text based ‘fighting fantasy book’. Here is the article about it, have a read through and then unpack the steps and learnings.

Warning this is an advanced prompt. For convenience the full prompt is in the end of the premium section.

The original prompt article for adventure books.

There is also an educational / history version rewrite of it. This is potentially a breakthrough use.

Did you miss these articles on our blog?

Our Events

Inside premium this week

• The Jagged Frontier, understanding the productivity gains and risks

• Choosing wisely, a model for picking AI candidate projects

• 2023 Generative AI Breakout year, plan for 2024

• Level Up player one.

• and a little prompting, Occam’s razor

• The Companies and Markets on Regulation of ‘Foundation Models’

Subscribe to Premium to read the rest.