The big tech | Issue 10

Welcome to Issue 10 of unfolding:ai, we are delighted to have you as a reader and as always hope you find this week’s issue insightful.

Are you getting the most out of the newsletter?

How to log on the the homepage, use the little ‘burger’ menu top right of corner of the unfoldingai newsletter home page. It will email you the login link

You can always read back issues on this page too, or check your referral count and rewards.

p.s bye dalle-2, you were first out the gate (april 2023), and almost instantly superseded by competitors, dalle-3, better look over your shoulder!

Best regards,

Paul, Co-founder (and newsletter editor)

AI big tech round-up

The AI space is dominated by the major ‘Big Tech’ providers, the majority of applications are built on these. In addition these platform providers are pushing AI into the core products, in essence moving the entry position away from some startup’s and innovators. Expect a Q4 of rollouts and updates.

Microsoft

The largest roll out of AI is taking place right now, Windows co-pilot is being rolled out to every windows 11 PC (in the consumer space). Here it is on my PC.

As of this week it is in pre-phase and pilot roll out, mainly replacing and updating the ‘bing’ button. Enterprise will be behind on this, due to the usual security, aged PC estate. There will be that period of time again where modern home technology is ahead of enterprise.

In addition to the ‘windows’ co-pilot, additional co-pilots are rolling out for every part of the Office and associated family, in Mid to Late November. It is being ‘heavily’ advertised and I know of a few enterprise customers who are on the beta versions. Feedback is good, but generally without any guidance on how to do things ‘differently’ there is a real danger that change will become ‘same task but quicker’, also risks and hallucinations are still in the product.

Personally I think, the improvements to excel look the most promising, bridging the gap to Power BI. Anything to make slide production easier is welcome, but also I fear more ‘death by slides’.

Licensing costs, (always unclear from micro$) are still opaque, but I would expect that there is a cost per seat uplift in the offering for all clients who want to take advantage above the most basic of use case.

There is also a ‘co-pilots’ lab / training site which is being heavily advertised/teased.

https://blogs.windows.com/windowsexperience/2023/09/26/the-most-personal-windows-11-experience-begins-rolling-out-today/

Google

Google last week released ‘bard for personal workspaces’. Basically anyone with a personal @gmail.com. It can be enabled on the website https://bard.google.com/ you then need to enable the workspace features to provide it access to your email / drive (if you want to).

The access to the workspace, holds promise, but at the moment bard is basically pretty poor. I found gmail search was better at finding emails.

No date on when this will be in full enterprise roll out, I have heard it is in testing in major enterprise workspace clients.

I expect a step change when bard moves to ‘gemini’, planned at the moment for 2024

Apple

They are spending about $1m per day on AI……. everything else is conjecture. I would expect to see ‘siri’ getting an uplift in Q1 2024, possibly in conjunction with further wellbeing / health machine learning.

OpenAi

OpenAI (chatGPT) have announced ‘multi-modal’, or to be more precise released it. The issue was available nVidia chips, gpt4v (for vision!) was always in the solution.

Multimodal - can ‘understand’ images, audio, and output both images and audio. Is a big step towards a general tool. Specifically in the space of ‘personal assistant’. This will probably kill off siri/alexa in the short term, apart from they have ‘privilege’ at a device level as an advantage. Image generation is ‘dalle-3’, which looks good from the curated examples.

The currently presented use cases (marketing) have no doubt been very carefully curated. Expect the desktop version around 23rd October, IOS / Mobile app is rolling out ‘now’. (As of writing It’s not in the UK)

It will be interesting to see if such a generalist tool, is of enterprise value, rather than consumer. The developer AI, to allow people to really get under how it works, will be later in the year.

Claude.ai & Amazon

Amazon are behind on the AI platform space, they have chosen to invest upto $4bn, though the initial amount is $800m (only!). Claude is a great AI, long context window for data, though perhaps midway between chatGPT 3.5 and chatGPT4, in overall ability.

Amazon and its AWS platform cloud model is stronger than microsoft Azure, though perhaps less ‘enterprise’, especially in retail. There is a long distrust over ‘data’ and ‘markets’ with Amazon and retailers.

If you haven’t tried claude as an alternative to chatGPT, it is worth an experiment.

We (unfold:ai) have just received developer claude access, so will be looking deeper into the tool set from a business productivity perspective.

Meta & open source

A quick mention on Meta (facebook) & open source. There is a lot of interesting enterprise scale models being released in the open source world. This is driving a lot of the frontier experimentation, multi-modal, Mixture of Experts, Synthesis pipelines. All of this is helping to build better models, train them quicker, cheaper.

One of the most interesting areas though is ‘agents’, which is the next logical step for AI, as it moves from being in the ‘thinking’ quadrant, into ‘sensing’ - data input, and then ‘acting’.

These event and autonomous agents will be the ‘next generation’ of AI, effectively combining events, processes, search and data transformation in to action.

What is Multi-Modal and Mixture of Experts (MOE)

Two terms are in current circulation this week, describing the continued growth of AI’s capabilities.

Multi-modal in the context of AI Language Models (LLM) like GPT-4 refers to the ability of the system to understand and generate different types of data, such as text, images, and sound, all within the same program. The idea of AI’s to do differing tasks is not new, neither is combining ‘vision’ and ‘language’, infact GPT 4 has always been multimodal, just a lack of available hardware prevented roll out.

Mixture of Experts (MOE) is ‘probably’ how chatGPT works. It refers to a system where multiple specialized "expert" models work together to provide a solution. Each expert is good at specific tasks. The system decides which expert to consult based on the problem at hand.

Advantages:

1. Specialization: Each expert is highly skilled in a specific area, leading to better results.

2. Efficiency: Can be faster as each expert only focuses on what they're good at.

Downsides:

1. Complexity: Managing multiple experts can be complex.

2. Cost: More computational resources are usually needed for multiple experts

Quick things to check out

Tesla robot.

Elon musk is still my ‘most likely evil genius to build killer robots by accident, whilst living in a volcano’ tech leader.

A prompt to try in claude.ai (and other LLMs)

Try using <> to help chatGPT to split its process and format. This is a side effect that LLM’s were trained on a large amount of content that included XML as a language, we can use that to our advantage. Also note, the prompt is built up to help claude reason to the end.

I would like you to suggest some fun activities for a halloween party, and then write an email invite.

<party attendees>
age range: 19 - 26
Income range: Early Career Professionals
Location: After work
Duration: 2 to 4 Hours
</party attendees>

<Activities>
Suggest three appropriate easy to do activities
</activities> 

<email invite>
Write a fun, professional short email invitiing people to attend and RSVP
</email>

Help us grow

We have a free referral scheme, where you can help our newsletter grow. Every referral that you make that accepts, adds to your ‘referral points’, and points makes prizes!

Thanks for sharing

What’s in our toolbox

Here is what we are currently using as our regular AI toolset.

On the main bench

chatGPT - General writing and ideation support

claude - Larger context data analysis

Midjourney - quick and easy splash / slide images

leonardo - more involved, higher quality, complex images

notion - Note and prompt organisation (and its inbuilt AI)

Canva - great for producing visuals, AI is still a bit sketchy but worth a go, more often though use stock images

Subscribe to Premium to read the rest.