HNHacker News
TopNewBestAskShowJobs

sthatipamala

3,538 karma · joined October 3, 2010

Sridatta Thatipamala

Reach me by email: sthatipamala@gmail.com

YC Badge: 0xa18d1d3604177ac278ed284fc251bb9330f31286

submissionscomments
sthatipamala··on Ask HN: What have you created that deserves a second chance on HN?
I hope this counts as original content I created...
sthatipamala··on Ask HN: What have you created that deserves a second chance on HN?
I'm a cohost of The Technium podcast. Please check us out!

It's a weekly podcast discussing the edge of technology and what we can build with it. Each week, my cohost Wil (https://news.ycombinator.com/user?id=iamwil) and I introduce a big idea in the future of computing and extrapolate the effect it will have on the world.

Wil and I started this show because we found that most tech podcasts were focused on career development or Big Tech drama. We wanted a show where we could be optimistic and excited about the future of software, especially things which were not mainstream.

Some of my favorite episodes:

- Smalltalk : https://www.youtube.com/watch?v=FZNqOFAhM8o

- Zig: https://www.youtube.com/watch?v=wie5YuzoUQI

- Generative AI models: https://www.youtube.com/watch?v=iOy-v2ah0Ms

We're in all the usual places:

- Youtube: https://www.youtube.com/@techniumpod

- WEBSITE: https://technium.transistor.fm/

- SPOTIFY: https://open.spotify.com/show/1ljTFMgTeRQJ69KRWAkBy7

sthatipamala··on On the dangers of stochastic parrots: Can language models be too big? (2021)
I am curious to see that trolley problem screenshot. I saw another screenshot where ChatGPT was coaxed into justifying gender pay differences by prompting it to generate hypothetical CSV or JSON data.

Basically you have to convince modern models to say bad stuff using clever hacks (compared to GPT-2 or even early GPT-3 where it would just spout straight-up hatred with the lightest touch).

That's very good progress and I'm sure there is more to come.

sthatipamala··on On the dangers of stochastic parrots: Can language models be too big? (2021)
This paper is the product of a failed model of AI safety, in which dedicated safety advocates act as a public ombudsman with an adversarial relationship with their employer. It's baffling to me why anyone thought that would be sustainable.

Compare this to something like RLHF[0] which has acheived far more for aligning models toward being polite and non-evil. (This is the technique that helps ChatGPT decline to answer questions like "how to make a bomb?")

There's still a lot of work to be done and the real progress will be made by researchers who implement systems in collaboration with their colleagues and employers.

[0] https://openai.com/blog/instruction-following/

sthatipamala··on Ask HN: What are some of the best podcasts for developers?
I'm a cohost of The Technium. Please check us out!

It's a weekly podcast discussing the edge of technology and what we can build with it. Each week, my cohost Wil (https://news.ycombinator.com/user?id=iamwil) and I introduce a big idea in the future of computing and extrapolate the effect it will have on the world.

Wil and I started this show because we found that most tech podcasts were focused on career development or Big Tech drama. We wanted a show where we could be optimistic and excited about the future of software, especially things which were not mainstream.

Some of my favorite episodes:

- Smalltalk - https://www.youtube.com/watch?v=FZNqOFAhM8o

- Zig - https://www.youtube.com/watch?v=wie5YuzoUQI

- Generative AI models - https://www.youtube.com/watch?v=iOy-v2ah0Ms

We're in all the usual places:

- Youtube: https://www.youtube.com/@techniumpod

- WEBSITE: https://technium.transistor.fm/

- SPOTIFY: https://open.spotify.com/show/1ljTFMgTeRQJ69KRWAkBy7

- APPLE PODCASTS: https://podcasts.apple.com/us/podcast/the-technium/id1608747...

sthatipamala··on Elastic.co lays off 13% of team
Where can I read about this? I’m not familiar with their history.
sthatipamala··on Semantic Search with SQLite
Any examples of this being combined with full text search to get a hybrid index?
sthatipamala··on Metaphor Systems: A search engine based on generative AI
That would help retrieve documents based on their contents. But you couldn’t query by a description of what kind of link it is.

So metaphor is able to translate the language of comments (“here are some thoughtful, technical blog posts about AI”) to the language of documents.

Disclaimer: I also work on a semantic search engine.

sthatipamala··on Performance reviews don’t assess performance
At big companies, leveling is sold as an objective/descriptive measure of one's skills. An engineer is at level L if and only if they meet the expectations at L. If you aren't at that level, it must be because you lack certain skills from the rubric.

What you are saying is true but it contradicts that model. It acknowledges that leveling has extrinsic criteria such as finite number of spots.

sthatipamala··on How to really know another person
I made a similar thing for the Lex Friedman podcast. It transcribes the audio and runs it through a summarization model. You can then search the entire corpus to find relevant snippets from episodes:

http://lexicon.lol

sthatipamala··on Launch HN: Iollo (YC S22) – At-home metabolomics test to extend healthy lifespan
Can I bill this to an HSA?
sthatipamala··on Why don’t we have a strategy?
People have said similar things about me as your Japanese peer. What follows is tough love for myself so take it for what its worth:

If no one listens to me when I'm right, that's my problem, not theirs.

Everyone has their own incentives and motivations. If my goal is to actually effect change, the onus is on me to translate my strategy into a plan that is executable by self-interested, messy human beings.

sthatipamala··on Kate Bush’s “Running Up That Hill” synth sounds
I have nothing to add on the synth/music aspect, but thank you for sharing your personal story. The idea of programming your mind to get through tough times really resonated with me.
sthatipamala··on FCC Closes Robocall Loophole
Here's the full text of the release. I don't know why they just don't just put it on the page itself... :

Media Contact: Will Wiquist will.wiquist@fcc.gov

For Immediate Release

FCC CLOSES ROBOCALL LOOPHOLE FCC Robocall Response Team Has Taken Enforcement Actions, Built Nationwide Partnerships, and Proposed Innovative New Policies to Combat Scam Robocalls -- WASHINGTON, June 30, 2022—Starting today certain small phone companies must comply with FCC rules to implement caller ID authentication tools on their networks, just as large voice service providers are required to since June 30, 2021. Today’s announcement is the latest in a series of actions by the FCC’s Robocall Response Team to cut off the flood of unwanted robocalls hitting consumers and business phone networks. These small phone companies are suspected of facilitating large numbers of illegal robocalls and, as a result, the FCC rolled back an extended caller ID authentication implementation timeline granted to them in its original 2020 rules.

“Each time I get a robocall it reminds me that we can’t stop looking for ways to stop these nuisance calls and the scams behind them,” said FCC Chairwoman Jessica Rosenworcel. “Our team is working to aggressively and creatively find ways to fight back. We will use every authority we have, and we will go to Congress for more. We will not let up.”

How We Got Here: In 2020, the FCC granted voice service providers with 100,000 or fewer subscriber lines an extension of STIR/SHAKEN* implementation requirements, consistent with the TRACED Act. However, since then evidence emerged that a subset of these small voice service providers were originating an increasing quantity of illegal robocalls. As a result, in 2021, the FCC unanimously voted to shorten the extension by a year.

Recent FCC investigations and reports from the Industry Traceback Group indicate that, since STIR/SHAKEN was widely implemented across the largest providers’ networks last year, robocallers have sought to maintain anonymity and avoid enforcement and blocking tools by routing or originating their call traffic on the networks of these largely IP-based* small providers that have not yet implemented STIR/SHAKEN. This has allowed robocalls to pass from these networks to terminating provider networks without carrying forward accurate and standardized caller ID/traceback information.

What’s New: Effective today, a problematic gap in FCC robocall rules closed, requiring non-facilities based small voice service providers* to implement STIR/SHAKEN caller ID authentication standards on their networks. These providers are now required to implement STIR/SHAKEN caller ID authentication standards on the IP portion of their networks.

The Bigger Picture: Under Chairwoman Rosenworcel’s leadership, the Robocall Response Team was created to serve as an FCC staff working group that pulls together expertise from across the agency to leverage the talents of enforcers, attorneys, policy makers, engineers, economists, and outreach experts to combat the unyielding menace of illegal spoofed, scam, robocalls.

This effort has resulted in: · record-breaking spoofing and robocall fines; · closing gateways used by international robocallers to reach Americans’ phones; · widespread implementation of STIR/SHAKEN caller ID authentication standards to help traceback illegal calls and improve blocking tools to protect consumers; · the signing of robocall investigation partnerships with the large majority of state Attorneys General; · and unprecedented policy proposals to combat the rising threat of bogus robotexts.

###

Appendix of frequently used terms: · STIR/SHAKEN Caller ID authentication: Caller ID authentication, based on so-called STIR/SHAKEN standards, provides a common information sharing language between networks to verify caller ID information which can be used by robocall blocking tools, FCC investigators, and by consumers trying to judge if an incoming call is likely legitimate or not. · Non-facilities-based voice service providers: A voice service provider is non-facilities based if it offers voice service to end users using connections that are not sold by the provider or its affiliates. Instead, their voice service is transmitted over another provider’s transmission service. · IP-based telephony: IP telephony is shorthand for Voice over Internet Protocol (VoIP), which is a technology that allows a user to make voice calls using a broadband Internet connection instead of a regular (or analog) phone line.

Media Relations: (202) 418-0500 / ASL: (844) 432-2275 / Twitter: @FCC / www.fcc.gov

This is an unofficial announcement of Commission action. Release of the full text of a Commission order constitutes official action. See MCI v. FCC, 515 F.2d 385 (D.C. Cir. 1974).

sthatipamala··on Master’s Degree in Computer Science
If lack of peer interaction was your main hangup, I think it has been addressed. The OMSCS Slack workspace has become a kind of 24/7 office hours, study group environment.

But there is still little to no interaction with the professors.

sthatipamala··on Master’s Degree in Computer Science
Based on what the OP wrote, Georgia Tech's OMSCS program would be worthwhile, especially if they are interested in the systems track. The UT Austin program is also modeled on GA Tech's program and looks especially good for its theoretical ML courses.

OMSCS has a lot of flexibility in being able to attenuate your course load, gives 6 years to complete the degree, and your peers are mostly working professionals. That's a big advantage of online-first programs.

sthatipamala··on Master’s Degree in Computer Science
I'm also a proud OMSCS alum! But "the opportunity to do some research and publish papers" is certainly not the default.

Can you describe how you were able to do that?

sthatipamala··on Show HN: I built a simulator for personal finance
You’re not too far away. A simple next step is to directly connect me with an advisor who uses ProjectionLab. I bet financial planning is a competitive market and advisors would be willing to use your tool as a unique selling point, or even pay you for the leads you supply.

Another bundling is to help users execute on their projections, similar to how https://www.helloplaybook.com does with tax advantaged investing.

More generally, I think there is an opportunity to market the product not as a tool (“ProjectionLab builds the best financial projections”) but as a solution to a problem (“ProjectionLab lets you retire early/plan to pay down your debt/buy a house”).

I know you’re doing this as a side project and maybe what I’m suggesting is too startup-y. Best of luck either way, it’s a really great product.

sthatipamala··on Show HN: I built a simulator for personal finance
Great work! This might be unpopular among the HN crowd, but I think you can monetize this further by bundling this with an advising service.

This idea of FIRE has been spreading beyond techies/Internet nerds and into other highly paid professionals like doctors. [0]. The main difference is that non-programmer professionals generally don't get a thrill out of min/max-ing financial projections.

Such people might be willing to pay a few hundred dollars a year for someone to generate this kind of analysis with them and simply execute on it each year. Similar to how people pay for CPAs. My employer actually pays Goldman Sachs to provide a similar service for its employees (no commission, fiduciary advisors) but their tools suck compared to this.

[0] For example https://www.whitecoatinvestor.com/, https://www.physicianonfire.com/

sthatipamala··on Transformers for software engineers
Thanks for this writeup! Software engineers should definitely learn about Transformers. They are a significant step-change in the advancement of deep learning. They've proven useful in multiple domains outside of natural language, including computer vision and audio.

This opens up the world of multi-modal models which represent concepts by considering multiple types of input. The most popular one has been OpenAI's CLIP [0]

If anyone's interested, I run a podcast and did an episode on Transformer models and their implications [1]. Check it out!

[0] https://openai.com/blog/clip/ [1] https://www.youtube.com/watch?v=Kb0II5DuDE0

sthatipamala··on Launch HN: Realize (YC W22) – Integrate brokerage accounts into your site or app
As a consumer, I’ve had significant difficulty getting my Schwab Equity Award Center balances to sync with apps like Wealthfront. The connection is always buggy and I have to reauthenticate all the time.

I wonder if this is the problem. If so, I’m glad you are working on this. Best of luck!

sthatipamala··on Ask HN: What do you do when your insight (& vision) scares you?
You should act on it! Your insight is important and valuable, but it’s just a starting point. Read some startup literature on validating problems. I recommend The Mom Test. You will want to talk to people to understand their existing behavior, without imposing your vision on them. Once you’ve elicited this from people, you will want to assess whether your conceptualization of the world was right. It might turn out that people were warring because it gives men something to feel manly about. Then your conflict resolution approach will have no effect. If you have validated that the problem exists, then the journey starts. You will have to figure out how you’re going to deliver your solution, how to promote it such that your audience will be receptive. Etc etc. But start with problem validation and be prepared to kill your precious vision at the altar of truth.

You seem like an idealist. So let me say two things:

1. don’t reject startup advice or business advice. Whatever your noble vision of the world, it will need a delivery and distribution mechanism to spread it. That will look like business and marketing. If you’re averse to it because it is “impure” you will only hinder your efforts.

2. It doesn’t matter if the world is not as it “should be”. Let’s say people are building swords because they just like fighting and feeling manly. It doesn’t matter that there is a theoretically better way to resolve conflict. Accept the world as it is and be humbled. It is you who was wrong, not every other person in the world.

sthatipamala··on Ask HN: What problem are you close to solving and how can we help?
Hmm... did you happen to work at Google? The concept and the UI seems very familiar.
sthatipamala··on Weird Languages
Strictly speaking about data analysis, tidyverse is way better. Even in adjacent spaces such as plotting and interactive data, tools like Shiny [0] are ahead of the Python ecosystem.

R is definitely weird language, not as useful for general purpose computing, and has far less marketshare. But that's kinda the point of pg's essay.

[0] https://shiny.rstudio.com/

sthatipamala··on Weird Languages
The whole tidyverse set of packages for R. Its a DSL for data analysis that makes heavy use of R macros. The vocabulary and mental model that it has about data transformation is superior to any other analysis package I've used.

Pandas/Numpy is close but R's macro/custom operators make everything much more seamless.

sthatipamala··on Show HN: I made a simulator for personal finance and financial independence
Yes but looking at Zillow that $2M Bay Area home would go for $250K in Houston TX. In fact, I can't even find homes small enough in Texas to compare some of the houses I see for sale in the Bay...
sthatipamala··on Show HN: I made a simulator for personal finance and financial independence
Just draw the user's attention to this and any derived/estimated values. It's impossible to model this correctly without very specific regional data.
sthatipamala··on Show HN: I made a simulator for personal finance and financial independence
This is super useful but the results were screwed up for me because of the default house maintenance estimate. For HCOL areas, 2.5% in yearly maintenance is HUGE.

I'm planning on buying a $2M house in the SF Bay Area and didn't realize this tool was allocating $50K a year just for upkeep. Just a point of feedback :)

sthatipamala··on A Facebook engineer abused access to user data to track down woman
If that was his plan, he picked the entirely wrong company to accomplish it. There are many layers of access control at Google that would make that impossible regardless of how highly placed he is. He'd be better off at a smaller company with less mature security tools.
sthatipamala··on Using GPT-3 for plain language incident root cause from logs
Polar (https://getpolarized.io/) has a GPT-3 based flash card generator from text highlights. It's available to premium subscribers.
← PreviousPage 2 of 9Next →