Thursday, April 27, 2023

AI is on the doorstep of big enterprises

Currently, companies concerned about information security are not allowing their employees to use AI-products such as ChatGPT. This makes perfect sense, as the data entered into ChatGPT will be used by OpenAI for training models and stored on their servers. At least that how it was until recently.



๐Ÿ” The other day, OpenAI rolled out the possibility to opt out from using your data in training new models, and they are planning to release a business subscription for "professionals who want to have more control over their data."
https://lnkd.in/dufUWU8v

๐Ÿ“ˆ This opens up the possibility of using ChatGPT with your company data, such as improving the language of a document or make a summary of it. Massively great.

๐Ÿ“Œ If you're not "enterprise-level concerned" about confidentiality but still want some level of privacy, opting out from sharing your data with OpenAI might be sufficient to get start using ChatGPT in your profession, check the fine print before going all-in. If this is not enough, the solution is just around the corner...

๐Ÿš€ The real revolution will come when the AI-model has access to all your internal data (documents, e-mails, chat conversations, etc.). Then, your own "ChatGPT" will provide answers based on your company data, allowing you to ask it, for example, to summarize the top 10 lessons learned from all projects in your company. This is getting closerโ€”a lot closer.

A few weeks ago, Microsoft demonstrated how to leverage ChatGPT's "brain" to interact with your enterprise data while respecting information protection controls. You can test it now.
https://lnkd.in/dzDcAaBk

๐ŸŽ“ As said many times, AI will revolutionize knowledge work, and it will happen very soon. My bet is that 2024 will be the year of re-skilling and up-skilling.

๐Ÿ’ญ What do you think? Have you started using ChatGPT professionally yet? What's holding you back? Has AI already improved your productivity or provided other benefits? How do you think about re-skilling your entire force of knowledge-workers?

Wednesday, April 26, 2023

Autonomous AI gives you autonomy

AI agents such as AutoGPT and AgentGPT, working autonomously towards goals you've set have recently gained attention.





Though AI acts autonomously, the key benefit is providing you more autonomy, e.g. say that you are planning a birthday party. You can decide to plan it all by yourself, delegate some tasks or the whole task to your AI birthday planner agent.

๐Ÿฅณ This is great because we (humans) like autonomy.

As a project manger, I often see proof of the power of autonomy. Assigning tasks to team members often yields half-hearted results. However, when the team members come up with the task themselves, they complete them in time with passion.

๐Ÿ“š This is behavior well understood in the literature. Autonomy is a key to human motivation and well-being is described in the Self-Determination Theory. The theory states that we have three psychological needs that drive our motivation and well-being:

๐Ÿ“Œ Autonomy: The need to feel in control of one's own actions and choices, and to act in harmony with one's values and interests.
๐Ÿ“Œ Competence: The need to feel effective in one's activities and to experience mastery over challenges.
๐Ÿ“Œ Relatedness: The need to feel connected and have a sense of belonging with others.

๐Ÿค– How can AI boost our motivation?
(1) Autonomy can be improved by AI-agents as described above. AI is also improving our perceived (2) competence - I can now write Python code and make art! (3) The impact on relatedness is less obvious to me and Iโ€™ll leave it out for now. But improving 2 out of 3 is amazing.

๐ŸŒŸ Following this lightweight reasoning, AI has the potential to increase our motivation and well-being. I like AI.

โ“What do you think? Are AI agents here to stay? Do you think that AI will improve our engagement? What does AI mean for you?

Looking forward to seeing research studies within this field, if you have please ping me ๐Ÿ™‚

Wednesday, April 19, 2023

What AI future do we want to have? Lex, Max what do you say?

The exponentially growing number of AI related posts in my feed are 99% focused on tools and current development. The sentiment is positive. Bullish. Gold rush. I love it โค๏ธ





For another perspective, I highly recommend the Lex Fridman podcast episode with the MIT professor Max Tegmark. In my opinion, Max is often drawn to the edges of problems (typical physicist mindset) and unafraid of big perspectives, listening to him sparks new thoughts in my head.
He has gained a lot of attention lately as one of the authors of the open letter urging for a 6-month pause in AI development, which is the major topic in the interview.

Max talks about our drive for improved efficiency, which has historically created wealth and automated away tedious and dangerous workโ€”great! But now we have a tool that can automate jobs people find meaningful and enjoyable, how do we tackle that?

Should we optimize everything as much as possible? What are we optimizing towards? Efficiency? Our happiness and fulfillment? Or something else? Could we fail and steer towards a unwanted goal?

Do we need to pause and think?
Thatโ€™s impossible!?

No, Max argues that we have successfully defined global guardrails before, e.g. for cloning of DNA, and it is possible to do again if we want to.

โ“What should we do? AI-Pause? or go all-in-AI? What do you think?

๐Ÿš€ This far, the ChatGPT revolution has been amazing for me. I feel empowered to take on new challenges since I have an assistant on my side. Awesome.

๐ŸŒŸ My dream scenario would be to leverage the AI assistant to remove tedious work, like helping me pay the parking ticket I just got ๐Ÿคฌ, and empower me to do new things e.g., creating graphic illustrations, programming Python scripts, and becoming an better writer. However, there are also scenarios that don't sound that appealing, like optimizing away face-to-face interaction with my colleagues. Where are we heading?

I think everything Max says boils down to taking the situation into our own hands, and steering towards the best possible outcome while trying to avoid the unwanted scenarios. He suggests a pause, there might be other ways.

๐Ÿ’กTo me, the 3-hour-long podcast episode served as a sobering break from the normal flow of information. But now, I'm ready to jump back into AI details, AI tools, and how to AI-optimize my work as a project leader โ€” and sometimes pause, reflect, and adjust the direction.

โ“What are your thoughts? Are you afraid of having your job automated away? Or do you feel AI-empowered ?

Wednesday, April 12, 2023

AI agents are here to help you

A recent release of a task-driven autonomous agent has caught a lot of attention lately (see comment). It gives a clear view on where we are heading, and it's amazing - AI is getting into the world and doing real jobs for us. This is how it works, you give an "AI agent":

- a name,
- a role, and
- a goal

and it will set up a task, execute it, review the outcome, and iterate until the goal is met, having internet access to search for data ๐Ÿคฏ



๐Ÿง‘โ€๐Ÿ’ป As you can see, this is quite different from ChatGPT, which does not have internet access and you have to be there and provide prompts for every step. If you tried programming with ChatGPT, you know what I'm talking about: prompt for code, get the code, try the code, paste the error message, get updated code, and iterate until the code runs. Now, the iteration part is automated.

๐ŸŒด For example, say that you want to plan for your summer vacation, just create an AI agent "TravelJoe" with the role of browsing the web for vacation trips and a goal to find a one-week vacation in Greece for 2 adults traveling from Stockholm. TravelJoe will now set up tasks, execute, review the outcome, and iterate until the goal is met. Exactly as you would do it yourself.

๐Ÿ•ต๏ธ Try AgentGPT, link below.

๐Ÿ’ผ Business will be changed

๐Ÿ•ต๏ธYou will use agents for many tasks. Order three easy-to-make dinners for four persons, I want all ingredients delivered at my doorstep tomorrow. Done.

๐Ÿš€ This is not a crazy dream about the distant future, it is happening now, and you can toy around with it to be a bit prepared when it hits us.

๐Ÿ‘ฉโ€๐Ÿ’ผ How would I use this in my role as a project manager? For example, I can ask it to give me a draft document based on data from similar projects and help me organize kick-offs, find meeting times, run meetings?
It might be a way around training company specific models?

โ“ Taking it one step further, am I a project management agent that can be replaced? Easily?

โ“What will happen to all comparison sites? Will all commercial websites also have a API tailored to being used by agents? (this might already be the case today?)

โ“ How would you use an AI agent professionally?

Looking forward to the next few iterations on this ๐Ÿ˜ƒ

Thanks for the link Simon Angeldorff ๐Ÿ‘

Thursday, April 6, 2023

โœจ Co-everything is the future

๐ŸŒŸ Thank you Microsoft for an amazing day, packed with inspiration delivered by great speakers. To summarize my take-away with four words:


โœจ Co-everything is the future.



๐Ÿ’ก One thought that kept coming back in my head is how organizations with full focus on their core business will find the energy to adopt and benefit from technologies that's being delivered at breakneck pace (monthly, weekly, daily)?


The gap will become enormous between those who "keep up" and those who don't, larger than before. If we look back, for example, at MS Word 6.0, released in 1993, and MS Word 97, which came four years later with new functionality like Clippy and the ability to draw pictures, one could get by quite well without learning the new features. Small gap.

๐Ÿš€ Now, employees who adopt the new technology might increase productivity in daily work by 50%. Large gap. This gap must be bridged somehow!

๐Ÿ”ง How do we solve this? Will it require huge change management efforts? Or will the new functionality be so streamlined and integrated that it's adopted automatically?

โ“ Any examples of large organizations using a structure to educate employees that matches the current speed? What is the best practice?

๐ŸŒฑ Sustainability - one last thing I hope for is that sustainability metrics will be highlighted with the same importance as financial metrics, this is an incredibly inspiring perspective neatly packaged by Yalle Elehu.

Thanks, I'm eager to explore and help ๐Ÿ˜€

Wednesday, April 5, 2023

How I 1x, 10x, and 100x my productivity using ChatGPT

 How I 1x, 10x, and 100x my productivity using ChatGPT


๐Ÿง  A few years ago, I created a mindmap based on 100 or so newsletters / podcasts from the genius David Stiernholm. It was a manual process: listen, take notes, and summarize on a mindmap. Many hours of well-invested time.

๐Ÿค– Now, I wanted to see if I could partner up with ChatGPT and set up a semi-automatic workflow from newsletter to mindmap. NOTE: I'm not a programmer; I cannot write any Python code myself.

๐ŸŽ‰ Yes, we managed to pull it off as a team, see the workflow at the end of the post.

๐Ÿ’ก What about the promised productivity gain?
1x: Generating the mind map manually took about 8-10 hours, which is approximately how long it took to develop and execute the workflow below.
10x: Without the help of ChatGPT, I estimate that the workflow work would take 80 hours. Starting with reading guides on how to install Python, learning to use APIs, and so on.
100x: Manually generating a mindmap for 100 new episodes (there are >600 in David's archive) would take 6-8 hours. Running the workflow below takes 5-10 minutes, depending on OpenAI usage, making the time from e-mail to mindmap 100x faster.

๐Ÿ” Four perspectives:
๐Ÿ“Œ I think using natural language for "coding" will enable broader adoption of automation tools, making us more productive. We will create our own apps tailored to our personal workflows.
๐Ÿ“Œ Similar workflows to the one above will soon be native in all email apps such as Outlook, see MS 365 Copilot announcement. Start thinking about how to use this capability already now.
๐Ÿ“Œ In doing this experiment, I learned a lot about using ChatGPT as an assistant, its limitations, and created an output that can be used to more efficiently consume the productivity related podcast episodes. However, if your end goal is to become more productive, the greatest challenge will still be to change or create new habits, which we haven't found a 100x solution for. Some things will still take timeโ€ฆ
๐Ÿ“Œ Without ChatGPT I would not have taken on this task, it is an enabler of epic proportions. I was in the drivers seat making the process very fulfilling.

๐Ÿค” Have you tried any ambitious projects together with ChatGPT? What are your impressions?
๐Ÿค”What is the first thing you will do when getting your hands on MS 365 Copilot (or similar)?

๐ŸŒGo ahead and explore

๐ŸงFor the curious reader, the following workflow was used:
1๏ธโƒฃ Copy 100 email newsletters to a folder
2๏ธโƒฃ Convert emails (eml-files) to txt-files
3๏ธโƒฃ Summarize all text-files (the API cost was 0.5 USD)
4๏ธโƒฃ Combine all text-files into one text-file
5๏ธโƒฃ Use ChatGPT4 to create a table*, categorizing all data in the text-file from the step above.

*A table is easily converted to a mindmap
**The main challange was the limitations in input and output length.


A 5-step process for your best ChatGPT prompts

Wanting to use ChatGPT more seriously, I quickly realized that prompt engineering is no joke. It is the key to unlocking ChatGPT.


๐Ÿ” Researching blogs for good prompt structures, I've arrived at a simple recipe for how to structure a good prompt:

1๏ธโƒฃ Ask ChatGPT to play a role
2๏ธโƒฃ Explain the task
3๏ธโƒฃ Clarify the objective
4๏ธโƒฃ Add context and constraints
5๏ธโƒฃ Specify the output

๐Ÿค” Well, does anyone follow it?

๐Ÿ’ป I went to GitHub for a repository of "Awesome ChatGPT Prompts" and asked ChatGPT to score 50 prompts on a scale from 1 to 5 for how well they follow the recipe above, higher score means better match. Results are in the picture.



๐Ÿ“Š The results are quite obvious, almost all of these โ€œawesome promptsโ€ follow the recipe.


What does it mean?
๐Ÿคท Can you achieve great results with a different recipe? Yes. Is the recipe perfect for every situation? No.
๐Ÿ‘ฉโ€๐Ÿ’ป If you are a prompt engineer, you will probably have a much better approach, but if you are like me - exploring and learning to use ChatGPT, it is a great boilerplate to radically improve the output.

โ“ What are your favorite prompting techniques? Would sharpening your prompting technique be of any value to your profession?

๐ŸŒ Go explore! You will be mostly amazed.