Jasenko Dizdarevic
@jasenko.bsky.social
Founder and CTO of Taldea ( https://www.taldea.ai/home ).
Engineer at heart, pushing the boundaries of native cloud-based technologies, especially in software testing domain. Feel free to get in touch!
Engineer at heart, pushing the boundaries of native cloud-based technologies, especially in software testing domain. Feel free to get in touch!
From everything I hear from the field, the dumbasses really *love* the LLMs. It finally makes them feel like they are doing the hard work. And they will explain it to you with a passion too :)
November 15, 2025 at 10:01 AM
From everything I hear from the field, the dumbasses really *love* the LLMs. It finally makes them feel like they are doing the hard work. And they will explain it to you with a passion too :)
^That is not an issue, if you generally keep to the spec. Smaller improvements are fine and will come to you naturally. Larger improvements go into the backlog for the next iteration. Then rinse and repeat, draft spec-> implementation->test->release.
July 29, 2025 at 7:49 PM
^That is not an issue, if you generally keep to the spec. Smaller improvements are fine and will come to you naturally. Larger improvements go into the backlog for the next iteration. Then rinse and repeat, draft spec-> implementation->test->release.
Write down, and in the case of UI, draw, the actual spec of what you want to release. Then implement just that. However be mindful of the following: you will still compromise on your spec during implementation (smaller improvements). Inevitably some designs will be changed, especially on the backend
July 29, 2025 at 7:49 PM
Write down, and in the case of UI, draw, the actual spec of what you want to release. Then implement just that. However be mindful of the following: you will still compromise on your spec during implementation (smaller improvements). Inevitably some designs will be changed, especially on the backend
(5/5) The results reported are supposedly achieved by increasing the computational depth in the "system 2" layer and removing the need for backpropagation as a means of reducing the computing space and time cost. It sounds exciting, but we will yet have to see independent verification.
July 29, 2025 at 6:52 AM
(5/5) The results reported are supposedly achieved by increasing the computational depth in the "system 2" layer and removing the need for backpropagation as a means of reducing the computing space and time cost. It sounds exciting, but we will yet have to see independent verification.
(4/5) An abstraction of an abstraction of human brain seems to be as the key idea, probably better known by the terms "System 1" and "System 2", as popularised by the late Daniel Kahneman.
July 29, 2025 at 6:52 AM
(4/5) An abstraction of an abstraction of human brain seems to be as the key idea, probably better known by the terms "System 1" and "System 2", as popularised by the late Daniel Kahneman.
(3/5) The authors claim the system they built uses only ~27M parameters and only ~1000 examples to surpass the conventional LLMs in solving Sudoku tasks.
It actually seems to imply a bit more than that - that the system *inferred* the rules of Sudoku solely based on those 1000 samples.
It actually seems to imply a bit more than that - that the system *inferred* the rules of Sudoku solely based on those 1000 samples.
July 29, 2025 at 6:52 AM
(3/5) The authors claim the system they built uses only ~27M parameters and only ~1000 examples to surpass the conventional LLMs in solving Sudoku tasks.
It actually seems to imply a bit more than that - that the system *inferred* the rules of Sudoku solely based on those 1000 samples.
It actually seems to imply a bit more than that - that the system *inferred* the rules of Sudoku solely based on those 1000 samples.
(2/5) If true though, it could be a first step in moving towards something more feasible than the waste of $$ and energy that the GenAI represents. The paper claims to achieve a much more efficient and Turing-complete alternative to CoT LLMs. They call their model 'Hierarchical Reasoning Model'
July 29, 2025 at 6:52 AM
(2/5) If true though, it could be a first step in moving towards something more feasible than the waste of $$ and energy that the GenAI represents. The paper claims to achieve a much more efficient and Turing-complete alternative to CoT LLMs. They call their model 'Hierarchical Reasoning Model'
Perhaps we should just shrug and accept that the modern consumer web applications of today will never reach the reliability of similar Microsoft MFC and .NET desktop applications from two decades ago. And that's saying a lot.
March 13, 2025 at 4:43 PM
Perhaps we should just shrug and accept that the modern consumer web applications of today will never reach the reliability of similar Microsoft MFC and .NET desktop applications from two decades ago. And that's saying a lot.
But we need to be honest that this is not the way to develop high quality software. The intrinsic motivation that was a given because of highly technical focus of the teams is gone and replaced by material incentives of the "ideas people".
March 13, 2025 at 4:43 PM
But we need to be honest that this is not the way to develop high quality software. The intrinsic motivation that was a given because of highly technical focus of the teams is gone and replaced by material incentives of the "ideas people".
... a lot of people are now switching careers by taking crash-courses in product management and "agile". On a human level I dont blame them - it is easier to do what such people perceive as relatively easy, but supposedly "valuable" busywork, than to grind night shifts at ER, for less money.
March 13, 2025 at 4:43 PM
... a lot of people are now switching careers by taking crash-courses in product management and "agile". On a human level I dont blame them - it is easier to do what such people perceive as relatively easy, but supposedly "valuable" busywork, than to grind night shifts at ER, for less money.
Some people took it a bit personal it seems :) I agree with you, we need skilled Product and Project Managers, and that also means having some skin in the game through previous technical background. The problem is, there are very few such left in the meanwhile....
March 13, 2025 at 4:43 PM
Some people took it a bit personal it seems :) I agree with you, we need skilled Product and Project Managers, and that also means having some skin in the game through previous technical background. The problem is, there are very few such left in the meanwhile....
There are good product managers, and there are strong project managers as well. Only, in my experience, they have been struggling to show due to many who have joined tech mostly for the promise of higher income and perhaps status. And as we know, monetary motivation is not an intrinsic one.
March 13, 2025 at 4:04 PM
There are good product managers, and there are strong project managers as well. Only, in my experience, they have been struggling to show due to many who have joined tech mostly for the promise of higher income and perhaps status. And as we know, monetary motivation is not an intrinsic one.
Every time a web app has tricked you into doing something you did not intend to do, and causing minor frustrations, you can thank the KPI-minded product managers. These people don´t really love or enjoy technology, they just want to optimise it to extract more "shareholder value".
March 13, 2025 at 4:04 PM
Every time a web app has tricked you into doing something you did not intend to do, and causing minor frustrations, you can thank the KPI-minded product managers. These people don´t really love or enjoy technology, they just want to optimise it to extract more "shareholder value".
I don´t downplay the role of people who do customer support, sales and also those who know how to do product and project management. But over especially the last 10 years, it is hard not connect the increasing entshittification of software services with the role KPI-based "optimisation" has played.
March 13, 2025 at 4:04 PM
I don´t downplay the role of people who do customer support, sales and also those who know how to do product and project management. But over especially the last 10 years, it is hard not connect the increasing entshittification of software services with the role KPI-based "optimisation" has played.
More like the other way around. People who joined tech because they heard there is money in it, but lack technical prowess, tend to add unnecessary bloat and layers of additional communication, due to their not-understanding of technical intricacies. There are exceptions of course, sometimes.
March 12, 2025 at 9:18 PM
More like the other way around. People who joined tech because they heard there is money in it, but lack technical prowess, tend to add unnecessary bloat and layers of additional communication, due to their not-understanding of technical intricacies. There are exceptions of course, sometimes.
No please, don't invent things I never said. Soft skills absolutely do matter as well as proven management techniques, such as OKRs or anything from Andy Grove's arsenal. What does not matter is yet another person on the payroll whose sole contribution is moving tickets on JIRA.
March 12, 2025 at 9:14 PM
No please, don't invent things I never said. Soft skills absolutely do matter as well as proven management techniques, such as OKRs or anything from Andy Grove's arsenal. What does not matter is yet another person on the payroll whose sole contribution is moving tickets on JIRA.
I never argued in the first place. Since you are in such an influential position it seems, please could you do a favour to a ton of people - and teach your coder PMs how to make the switch-off-Gemini function in Google Workspace actually work? No sarcasm here, I really mean it.
March 12, 2025 at 9:10 PM
I never argued in the first place. Since you are in such an influential position it seems, please could you do a favour to a ton of people - and teach your coder PMs how to make the switch-off-Gemini function in Google Workspace actually work? No sarcasm here, I really mean it.
Please do keep pushing the envelope, your team absolutely delivers a ton of value!
March 12, 2025 at 8:34 PM
Please do keep pushing the envelope, your team absolutely delivers a ton of value!
Thanks for confirming my point! We'd be delighted to learn what their actual contributions on an ultra-technical project like Linux are... it's a bit of mystery ;)
March 12, 2025 at 6:28 PM
Thanks for confirming my point! We'd be delighted to learn what their actual contributions on an ultra-technical project like Linux are... it's a bit of mystery ;)
ArgoCD is nothing short of a fantastic product.
March 12, 2025 at 6:24 PM
ArgoCD is nothing short of a fantastic product.
...and I am sorry but the blogs you linked do not seem to prove your point at all, I am very confused as to why would you be linking them here. I guess I have to much visible work on my hands to care.
March 12, 2025 at 6:22 PM
...and I am sorry but the blogs you linked do not seem to prove your point at all, I am very confused as to why would you be linking them here. I guess I have to much visible work on my hands to care.
No, it's absolutely not the same. Playing with dark patterns to trick users into clicking buttons they did not mean to click is not comparable to real work done by engineers. I've no idea what are you trying to prove. Just show me a PM contribution in the Linux repo already ;)
March 12, 2025 at 6:18 PM
No, it's absolutely not the same. Playing with dark patterns to trick users into clicking buttons they did not mean to click is not comparable to real work done by engineers. I've no idea what are you trying to prove. Just show me a PM contribution in the Linux repo already ;)
Erm... Release Management != Project Management. Please don't conflate the two.
March 12, 2025 at 4:36 PM
Erm... Release Management != Project Management. Please don't conflate the two.
But it's not the PMs which are merging the code into the Linux codebase, are they? Nor are they contributing to the project otherwise? To declare Linus the PM for the project is ignorant in the best case, he is more of a chief architect, though he'd scoff at that sort of corporate title...
March 12, 2025 at 4:04 PM
But it's not the PMs which are merging the code into the Linux codebase, are they? Nor are they contributing to the project otherwise? To declare Linus the PM for the project is ignorant in the best case, he is more of a chief architect, though he'd scoff at that sort of corporate title...