23 | !Forbes 30u30 | Making the AI cyberspace personalized | Co-founder and CTO @getalchemyst | @ai4bharat alum | Chief Automation Officer for @ritikaadass
> Got denied leaves for offcampus by college TPM because the same company came for oncampus as well, and my TPM got to know it.
> Was told basically "CGPA dekhi hai teri?"
> surprised_pikachu.jpeg
> Got shadowbanned in placements
> Got a 10+ LPA job from oncampus because recruiters got to know me from a hackathon
> Left offer, one of my friends got in my stead, me happy
> Got $100k+ USD WFH offer, left that to join another promising startup (WFH)
> Punished by being forced to stay an extra sem
> TPM made an example out of me by sending a mail (can only DM SS for obvious reasons)
> Became part of my batch lore, instant popularity - asked out even on LinkedIn lmao (but I already won in life with my gf)
> dab_while_crying_internally.jpeg
> handled full time job with extra penalty courses and B Tech Final Year Project
> Worked on Data conflicts and hallucinations in LLMs as my BTech thesis
> Started building @getalchemyst from college room, with @uttaranxnayak
> Resigned from aforementioned startup job
> Advisor fac didn't know what I was working on, gave me less marks
> Left my job, raised preseed funds
> Btech thesis got published in Springer
> happy_happy.jpeg
> Fac indirectly apologized later on
> never_give_up.jpeg
people really have lost their basic sense of manners while speaking to elders, especially teachers.
i agree some teachers might not be good, doesn't give me the right to be rude to them, they're still teaching people or have taught people, and people should respect that
since ive officially become an ai infra guy, i have a question that's been bothering me for a while:
given how important agentic memory has become these days, what are the chances that memory does not only remain as a feature, but ultimately becomes a part of infra in the upcoming future?
like imagine:
the year is 2040.
the world has autonomous agents, companies, research labs, software systems, robots and what not.
all of them would need some form of a memory layer right?
What kind of X account am I to you?
1. Non-toxic
2. Meaningful
3. Boring
4. Annoying (in a good sense)
5. Delete krde bsdk
6. Good guy
7. I hate you
8. Funny
9. Popular
10. Always on your timeline
11. Scared to DM
12. Toxic
What kind of X account am I to you?
1. Non-toxic
2. Meaningful
3. Boring
4. Annoying (in a good sense)
5. Delete krde bsdk
6. Good guy
7. I hate you
8. Funny
9. Popular
10. Always on your timeline
11. Scared to DM
12. Toxic
I had 6k rejections, not counting the ones who ghosted me (which might be upto 2x).
Landed into 250 interviews, closed 17 offers.
This was before LLMs became good enough to write applications.
@ni5arga@cbseindia29 good morning CBSE, you said you used scanners to scan these copies,
now since the copies are out to the public view, do you mind explaining
which copies when scanned through a scanner, have a drop shadow? and these 3 folds?
did you really use scanners?
Apologies Aravind, I digress. This is pure damage control wrapped in verbose signalling. Vague terms like "expert cybersecurity professionals" don't help. Also, they could've tagged Nisarga (@ni5arga), Siddhant and others.
For reference, you can see vulnerability reports and RCEs by companies like @vercel, @Cloudflare and even by some YC backed startups.
Scale isn't the issue, priorities are.
Good to hear this. The messaging and informing the public on what they are doing in improving the cybersecurity itself is reassuring. Appreciating the ethical hackers and the public for pointing out the issues and the acknowledgement is a welcome change.
@HarshitNay80531@getalchemyst Publishing benchmarks for the same soon.
Regarding latency, it can get a 8s chunk done in 2.5s without GPU. Would work well for use cases like Wispr Flow, but fully local.
https://t.co/toR9wQ4qDe
Just open-sourced a few on-device STT models with Indic Language Support (8 so far on the OSS version).
These are v0 checkpoints (like, 2 weeks old) of what we're cooking at Alchemyst (which will have 22+ languages supported).
Feel free to try it out ^_^
We have been closely monitoring the vulnerabilities in the OnMark portal of our service provider that are being flagged in the public domain. An expert team of cybersecurity professionals has been deployed over the last few days from across various arms of the government as well as the IITs to fortify these systems, including taking them over to a more secure set up. The identified vulnerabilities have been contained, and other exploitable weaknesses are being ruled out.
We are grateful to all alert citizens and ethical hackers pointing out such weaknesses, and have gotten in touch with some of them directly.
We request any others to reach out to our security teams at [email protected] for any further inputs.
@dpradhanbjp@EduMinOfIndia@sanjayjavin@PIB_India@PIB_Edu@DDNewslive@AkashvaniAIR@airnewsalerts@PTI_News
Atp, India's whole EdTech stack needs to be revamped.
Why can't upgrade these software systems? I can bet the majority of SDE2 folks in this country can write safer code than this.
@IndianCERT please don't go via the red tape route again. Hold week-long hackathons to get the basic MVPs correct. I can bet you'll get better MVPs than in a week, compared to what these mutts have delivered over years.
Got response from @IndianCERT within an hour.
Top speed.
Indian govt need to employ better technical people to update the system otherwise malicious actor would do a lot of damage!