AI Whisperer, Grumpy One
Joined March 2010
- Tweets7.2K
- Following282
- Followers321
- Likes1.6K
This is a no brainer. Any software engineer who does not get this is digging their own grave of their career.
Finally got my copy of @welchlabs illustrated guide to AI. Can’t wait for Vol 2.
Beautiful book and really good learning material
PSA do not buy @urbancompany_UC products, the after sales service process treats you like a criminal before sending someone to check because of them footing the bill on warranty.
Fell for their marketing of better service than aquaguard. Wish I had bought atomberg instead.
Karan K retweeted
GPT-5.5 released
teapawt.org/OpenAI/status/20473765…
BillDesk and Tally are the worst products in India which thrive purely due to self inflicted choices of Indian companies of paying for cheapest with no thought for users. Distribution is king.
"How Google humiliates Indian Developers with Google Play BillDesk Verification"
Scary to read stories of people having to pay bribes of 50K to 2 lakhs to BillDesk. @RBI what is this?
reddit.com/r/india/comments/…
I have mostly stopped using openai dots as of now. Its just not very useful in its current avatar. Will revisit it after the next update.
Very different experience compared to my cofounders who have used grokbot daily and is tightly integrates in their workflows.
Karan K retweeted
Some stats on #php2shell. In the 10 hours before it was publicly announced, across the @Cloudflare network we saw thousands of attempts to exploit it. That jumped to tens of thousands in the 45 minutes post announcement, hundreds of thousands within 2 hours, and millions now.
I just read @GergelyOrosz ‘s newsletter about AI budgets per dev. Quite insightful on the range of views. There is one additional view for budgeting for AI that is not mentioned but I feel is worth discussing about.
The spend on infra specific for AI agents is something that should be categorized and tracked. We have started doing it internally and my instinct tells me the budget for it will be more than all existing infra combined for humans. This is excluding the cost of ai tokens itself.
Building the surrounding infra for efficient fast agentic access is still in its infancy and its going to be as important as “cloud spend” budget is for businesses today.
Karan K retweeted
So in fact we actually can get robots, flying cars, superintelligence, fast internet on planes, space travel, space colonies, orbital data centers, unlimited energy, likely cures for nearly every disease, and more all in our life times. Pretty great.
Karan K retweeted
I don't know, man. Sometime in the last year or so I really started to get this unsettling feeling of living in the future, and imo we're still at the indrawn-breath moment before all the things being previewed happen in earnest.
I have been using openai dots for past week while traveling on vacation. One of the things I noticed is that I have to be prescriptive to it on how to do things. Just like codex.
It does have a good memory but it still needs to be told to do more work before coming back with an answer. Frequently it will do the naive ai research and eagerly get back with a suboptimal answer/solution.
After prodding it to do more work it will go back and give better output. Simple things like searching for hotels/restaurants with some specific criteria.
Sadly I do not fully trust its judgement and I still have to do additional work on top to validate it.
It still has not gotten over the hump of “I trust by default” feeling that is crucial in these cases.
One of the best use-cases for @OpenAI dots would have been the hundred million forms that require filling over and over again with the same data but for different providers like visas, arrival forms, web check ins, bookings/reservations etc.
But due to the unpredictability of the scoping of access to this sensitive PII - I do not trust it to fill it only when required.
It can use all this information anywhere incl malicious sites that can just ask the AI to fill it. The entire UX is a single chat with no scoping. Example - my passport info from few days ago can or cannot be used for some form filling today.
What will happen a year or two from now? My passport will expire in a few years - will it use new one? How do i delete the info of old one?
Too many questions and the behaviour being scoped to one global chat is a big hindrance in adopting the dots format.
I think grok @bot structure is much better with multi bot support with each one being scoped to a particular thing and them coordinating with each other. It also makes it much easier for the bots to be integrated for work vs personal tasks.
Karan K retweeted
The only generation that knows how computers work is also the most relaxed about AI:
Dealing with dot bot has been very hit or miss. The feeling of the product has been a very hit or miss depending on situation.
Very buggy - does not respond to calls. Stops tracking task mid convo etc etc
It does not “click” with the user as well as grokbot nor did it have the solid experience that codex did when it finally became great experience back in Dec 2025.
I am not sure that dot will be winner in this category yet. But its been only 3 days! Lets see how it goes.