Madainn mhat a h-uile diune

Good morning everyone

'S e Cory an t-ainm a th' orm

My name is Cory

Tha mi nam innleadar bathar bog aig Voze

I am a software engineer at Voze

Tha mi a bhruidhinn ribh ann an Gàidhlig

I am speaking to you in Scottish Gaelic

'S e canan a th' ann a th' air a bhruidhinn le  nas lugha na caogad 's a h-ochd mìle daoine air feadh an t-saoghail bho 2022

It is a language that is spoken by less than 58,000 people world wide as of 2022

Tha sin a’ ciallachadh nach eil duine agam airson a chleachdadh

That means I have no one to practice with

ach sibh

Except you

tha mi a' dol a bhruidhinn Ribh an diugh mu A.I.

I'm going to talk with you today about A.I.

Ach nì mi e ann am Beurla, air do shon

But I'll do it in English, for your sakes

Is this talk for you?

• Stans

• Luds

• Skeptics

• Utopianists

• Scared

• Ignorant

• Hubrists

...to one it is given to create the things of art, and to another to judge what measure of harm and of profit they have for those that shall employ them.’

- From Plato’s Phaedrus

Anthropic CEO says that in 3 to 6 months, AI will be writing 90% of the code software developers were in charge of

March 13, 2025

nypost.com/2025/05/31/business/ceo-warns-ai-could-wipe-out-1-in-2-white-collar-jobs-in-next-five-years/

Anthropic CEO Hilariously Claims AI Will Double Human Lifespans Within a Decade

January 24, 2025

gizmodo.com/anthropic-ceo-hilariously-claims-ai-will-double-human-lifespans-within-a-decade-2000554601

50% of entry-level white-collar jobs automated in 5 years: In May 2025

May 2025

www.businessinsider.com/anthropic-ceo-ai-90-percent-code-3-to-6-months-2025-3

Nvidia CEO Jensen Huang says Anthropic CEO is very wrong, and on almost everything he said about AI:

Jun 14, 2025

timesofindia.indiatimes.com/technology/tech-news/nvidia-ceo-jensen-huang-says-anthropic-ceo-is-very-wrong-and-on-almost-everything-he-said-about-ai-dont-do-it-in-a-dark-room-and-tell-me-/articleshow/121851471.cms

the most consequential technology of all time, not just our time.

25 Mar 2025

https://www.endofmiles.net/nvidia-ceo-ai-is-the-most-consequential-technology-of-all-time/

Hyperbole

Smaller proprietary training datasets are driving progress, not giant open datasets: Training data from humans using LLMs and giving them feedback –– “Reinforcement Learning from Human Feedback” or RLHF –– is becoming a more significant share of signal.

Niche applications, not general intelligence, is making gains: We’ve been waiting for AGI for quite some time, but it’s always around the corner. Meanwhile, vertical-specific applications are making major gains as developers learn to apply LLMs and their techniques to narrow domains.

https://www.dbreunig.com/2024/12/05/why-llms-are-hitting-a-wall.html

Big gains are being made in efficiency, not intelligence: ...Amongst the leading model builders, while quality has plateaued, efficiency has skyrocketed.

All the cool kids are suddenly all-in on “AgenticAI.” My inner-skeptic says this is a key indicator general LLM progress truly has stalled; chatbots can’t solve the problems we give them in one shot, so let’s give them several. Perhaps agentic models will yield better results –– by many metrics o1 already has –– but they’ll be even slower and pricier.

As designers, developers, and other builders learn how to apply AI cogs selectively –– delivering ‘quiet’ AI features –– we’ll get improved existing tools and wholly new ones.

LLMs have hit a wall. Now begins the slow climb upward.

What is the future of A.I.

The End.

All Transformational Technology Follows a pattern of progression

Arvind Narayanan & Sayash Kapoor 

A.I. as Normal Technology

An alternative to the vision of A.I. as a potential superintelligence

[D]espite the obvious differences between AI and past technologies, they are sufficiently similar that we should expect well-established patterns, such as diffusion theory to apply to AI, in the absence of specific evidence to the contrary.

AI as Normal Technology; Arvind Narayanan and Sayash Kapoor; 2025

https://kfai-documents.s3.amazonaws.com/documents/2b27e794d6/AI-as-Normal-Technology---Narayanan---Kapoor.pdf

Diffusion of innovations

1. the innovation itself
2. adopters
3. communication channels
4. time (usually measured in decades)
5. a social system

Diffusion of Innovation Model

Transformative technology Progression

Printing Press

100 yr

200 yr

300 yr

Electricity

50 yr

100 yr

150 yr

Industrial Revolution

25 yr

50 yr

75 yr

Internet

5 yr

10 yr

15 yr

A.I.

6 mo

1 yr

1.5 yr

Not all technologies with potential diffuse.

• Betamax
• Laserdisc
• WebOS
• Segway
• Apple Newton
• Flash
• Hydrogen Fuel Cells
• Cold Fusion
• Eugenics
• Communism

Text

A.I. will replace your job

those that Won't adopt now will be replaced by those that will

A.I. Only produces Slop

The "Bitter Lesson"

The biggest lesson that can be read from 70 years of AI research is that general methods that leverage computation are ultimately the most effective, and by a large margin.

This is a valuable observation about methods, but it is often misinterpreted to encompass application development.

In the context of AI-based product development, the bitter lesson has never been even close to true.

AI diffusion in safety-critical areas is slow

In this broad set of domains, AI diffusion lags decades behind innovation.

Safety

A good example is Epic’s sepsis prediction tool which, despite having seemingly high accuracy when internally validated, performed far worse in hospitals, missing two thirds of sepsis cases and overwhelming physicians with false alerts.

Diffusion of Innovation Model

The Speed of Human

• Society won't change faster than society allows itself to change
• Some change is generational
• Less consequential changes can be comparatively rapid.
• Massive societal changes must overcome massive push back whether or not the pushback is warranted.

in August 2024, 40% of U.S. adults used generative AI. But, because most people used it infrequently, this only translated to 0.5%-3.5% of work hours (and a 0.125-0.875 percentage point increase in labor productivity)

Is individual adoption even as fast as it appears?

 Diffusion occurs over decades,
not years

The generality of LLMs allowed developers to fool themselves into thinking that they were exempt from the need to find a product-market fit, as if prompting a model to perform a task is a replacement for carefully designed products or features.

A.i. generated code will lead to a diminished skillset and ability to write code ourselves

If men learn [writing], it will implant forgetfulness in their souls. They will cease to exercise memory because they rely on that which is written, calling things to remembrance no longer from within themselves, but by means of external marks.

- Plato

Forcing the issue

• Diffusion of new technologies never comes by way of mandate.
   
• At best, mandates allow individuals to try something sooner than they otherwise would have.
   
• Unintended and unanticipated consequences can engender resentment and revolt when new technologies are forced upon individuals before they are ready to receive it. 

Forcing the issue

A.I. radically lowers the barrier to entry

Entry Level (2019)

Beginner Level (2019)

Mid Level (2019)

Senior Level (2019)

A.I. radically lowers the barrier to entry

What an experience Software Engineer can do

What an experience Software Engineer can do with A.I. Assistance

What CEOs think they can do with A.I. assistance

The universe of things that can be done

Automate your self out of your current job...

A.I. may or may not be a part of how you process out, or what you transition into

¯\_(ツ)_/¯ 

• Early adopters are better positioned to command more money, and more opportunities should adoption scale.
   
• Enthusiastic early adopters are heeded more than nay-saying laggards when it comes to identifying technological limitations.
   
• In organizations push or mandating A.I. use, early adopters wield outsized influence over organizational direction.

Benefits of early Adoption

Architecture

The relationship between software and its environment, and the management of the boundary between them, is what architecture is all about.

https://www.linkedin.com/pulse/complexity-considered-harmful-barry-o-reilly-ylzwf/

- Complexity Considered Harmful? Oh starling, you can’t say that! - Barry O'Reilly

Architecture

Almost definitionally, A.I. cannot provide architecture without having all the context needed about the environment the software is build in. This is not possible today. There is currently no foreseeable path to be able to accomplish that in the future.

Architecture

No 🤓 🕵️‍♂️

• Write very little code, not no code
• Still responsible for setting constraints which are informed by architecture from the engineer(s)
• Architecture is defined by engineers and embedded in the constraints.

Persistent Challenges

1. Cost - Compute & Context size
2. Reliability - Inherent non-determinism
3. Privacy
4. Safety & Security - Misuse/Bad actors
5. User Experience - Constraints of current media

www.cs.utexas.edu/~EWD/transcriptions/EWD06xx/EWD667.html

On the foolishness of "natural language programming".

Since the early days of automatic computing we have had people that have felt it as a shortcoming that programming required the care and accuracy that is characteristic for the use of any formal symbolism.

some still seem to equate "the ease of programming" with the ease of making undetected mistakes.

The virtue of formal texts is that their manipulations, in order to be legitimate, need to satisfy only a few simple rules; they are, when you come to think of it, an amazingly effective tool for ruling out all sorts of nonsense that, when we use our native tongues, are almost impossible to avoid.

 

Instead of regarding the obligation to use formal symbols as a burden, we should regard the convenience of using them as a privilege: thanks to them, school children can learn to do what in earlier days only genius could achieve.

www.cs.utexas.edu/~EWD/transcriptions/EWD06xx/EWD667.html

When all is said and told, the "naturalness" with which we use our native tongues boils down to the ease with which we can use them for making statements the nonsense of which is not obvious.

I suspect that machines to be programmed in our native tongues —be it Dutch, English, American, French, German, or Swahili— are as damned difficult to make as they would be to use.

We know in the meantime that the choice of an interface is not just a division of (a fixed amount of) labour, because the work involved in co-operating and communicating across the interface has to be added. We know in the meantime —from sobering experience, I may add— that a change of interface can easily increase at both sides of the fence the amount of work to be done (even drastically so). Hence the increased preference for what are now called "narrow interfaces". Therefore, although changing to communication between machine and man conducted in the latter's native tongue would greatly increase the machine's burden, we have to challenge the assumption that this would simplify man's life.

How much work goes into "setting up" AI to perform accurately. Is is more or less work than writing code -- especially with decent and narrow code-completion