Mimansa Jaiswal

I am disappointed in the AI discourse steveklabnik.com/writing/i-am...

Meta Addressing bias in LLMs It's well-known that all leading LLMs have had issues with bias-specifically, they historically have leaned left when it comes to debated political and social topics. This is due to the types of training data available on the internet. Our goal is to remove bias from our Al models and to make sure that Llama can understand and articulate both sides of a contentious issue. As part of this work, we're continuing to make Llama more responsive so that it answers questions, can respond to a variety of different viewpoints without passing judgment, and doesn't favor some views over others. We have made improvements on these efforts with this release—Llama 4 performs significantly better than Llama 3 and is comparable to Grok:

• Llama 4 refuses less on debated political and social topics overall (from 7% in Lama 3.3 to below 2%). • Llama 4 is dramatically more balanced with which prompts it refuses to respond to (the proportion of unequal response refusals is now less than 1% on a set of debated topical questions). • Our testing shows that Llama 4 responds with strong political lean at a rate comparable to Grok (and at half of the rate of Llama 3.3) on a contentious set of political or social topics. While we are making progress, we know we have more work to do and will continue to drive this rate further down. We're proud of this progress to date and remain committed to our goal of eliminating overall bias in our models.

Meta introduced Llama 4 models and added this section near the very bottom of the announcement 😬

“[LLMs] historically have leaned left when it comes to debated political and social topics.”

ai.meta.com/blog/llama-4...

"We train our LLMs on art and literature and educational materials, and for some reason they keep turning out progressive."

Have work on the actionable impact of interpretability findings? Consider submitting to our Actionable Interpretability workshop at ICML! See below for more info.

Website: actionable-interpretability.github.io
Deadline: May 9

𝐇𝐨𝐰 𝐜𝐚𝐧 𝐰𝐞 𝐩𝐞𝐫𝐟𝐞𝐜𝐭𝐥𝐲 𝐞𝐫𝐚𝐬𝐞 𝐜𝐨𝐧𝐜𝐞𝐩𝐭𝐬 𝐟𝐫𝐨𝐦 𝐋𝐋𝐌𝐬?

Our method, Perfect Erasure Functions (PEF), erases concepts perfectly from LLM representations. We analytically derive PEF w/o parameter estimation. PEFs achieve pareto optimal erasure-utility tradeoff backed w/ theoretical guarantees. #AISTATS2025 🧵

New paper from our team @GoogleDeepMind!

🚨 We've put LLMs to the test as writing co-pilots – how good are they really at helping us write? LLMs are increasingly used for open-ended tasks like writing assistance, but how do we assess their effectiveness? 🤔

arxiv.org/pdf/2503.19711

I can help! (Might end up being by 4am tomorrow if that works for you)

pre aca you would specifically avoid being diagnosed or seeking treatment if you didn't have health insurance to prevent it from making it impossible for you to get health insurance. when you bought health insurance after doing this you committed fraud.

i did this.

Some of his readers have asked Mike Masnick @mmasnick.bsky.social why his technology news site, Tech Dirt, has been covering politics so intensely lately. www.techdirt.com/2025/03/04/w...

I cannot recommend Mike's reply enough. It's exactly what readers need to hear, what journalists need to do.

Neat visualization that came up in the ARBOR project: this shows DeepSeek "thinking" about a question, and color is the probability that, if it exited thinking, it would give the right answer. (Here yellow means correct.)

I think I did 6 hours a day for a week for leetcode. Looking back (I do mention it in the post), but learning the exact code for binary search, graph components, and coding attention blocks -- they were used so often and memorizing them (rather than writing them on the fly) helped down the line.

The Allen Institute for AI

Come work with me!
We are looking to bring on more top talent to our language modeling workstream at @ai2.bsky.social building the open ecosystem. We are hiring:
* Research scientists
* Senior research engineers
* Post docs (Young investigators)
* Pre docs

job-boards.greenhouse.io/thealleninst...

I am glad! Suggestions and comments are always welcome :)

LLM (ML) Job Interviews - Resources A collection of all the resources I used to prepare for my ML/LLM research science/engineering focused interviews in Fall 2024.

Resources as a notion database: mimansajaiswal-embedded-dbs.notion.site/19223941af7...

You can contribute resources at: mimansajaiswal-embedded-dbs.notion.site/19e23941af7...

Link to the resources post: mimansajaiswal.github.io/posts/llm-m...

LLM (ML) Job Interviews (Fall 2024) - Process A retelling of my experience interviewing for ML/LLM research science/engineering focused roles in Fall 2024.

You can read more about the process here: mimansajaiswal.github.io/posts/llm-m...

OCR'ed text from screenshot of top of post: LLM (ML) Job Interviews (Fall 2024) - Process A retelling of my experience interviewing for ML/LLM research science/engineering focused roles in Fall 2024. This post has two parts: Job Search Mechanics (including context, applying, and industry information), which you can continue reading below, and, Preparation Material and Overview of Questions, which you can read at LLM (ML) Job Interviews - Resources Disclaimer Last Updated: Dec 24, 2024 This is the process I used, which may work differently for you depending on your circumstances. I am writing this in December 2024, and the process occurred during Fall 2024. Given how rapidly the field of LLMs evolves, this information might become outdated quickly, but the general principles should remain relevant. (more...) Read at: https://mimansajaiswal.github.io/posts/llm-ml-job-interviews-fall-2024-process/

I interviewed for LLM/ML research scientist/engineering positions last Fall. Over 200 applications, 100 interviews, many rejections & some offers later, I decided to write the process down, along with the resources I used.

Links to the process & resources in the following tweets

Towards Integrating Personal Knowledge into Test-Time Predictions Machine learning (ML) models can make decisions based on large amounts of data, but they can be missing personal knowledge available to human users about whom predictions are made. For example, a mode...

Obsessed with the work coming out of Finale Doshi-Velez's group; they don't just take the limits of the real world for ML deployment seriously but instead turn it into new algorithmic ideas
arxiv.org/abs/2406.08636

What softwares do I actually use on my Mac as a software enthusiast? • Mimansa Jaiswal With several years of using a Mac, it took me time to settle down on a set of apps and softwares that I can heartily recommend. I try almost 30 a month, but end up using around 20 in total for everyth...

Ooh, I'd love to read this. I have mine here: mimansajaiswal.github.io/posts/mac-so... . I am currently testing out Atuin. I typically don't add new tools to my workflow until I've used them for at least a year or feel confident that I will continue using them long-term.

Definitely sponsored by Big Cupcake. Did you know Big Cupcake and Big Muffin are different AND competing corporations? 😃

YouTube video by The Kelly Clarkson Show Anne Hathaway Shares Genius Cupcake Hack

Reminds me of this solution 🤣: youtu.be/MTQUrUbb8vo?...

The entire archive of CDC datasets can be found here.

HUGE shoutout to data archivists- this work is important 👏🙌🏻

archive.org/details/2025...

Ai2 ScholarQA logo

Can AI really help with literature reviews? 🧐
Meet Ai2 ScholarQA, an experimental solution that allows you to ask questions that require multiple scientific papers to answer. It gives more in-depth and contextual answers with table comparisons and expandable sections 💡
Try it now: scholarqa.allen.ai

Political News for Normal People. Your essential guide to the shock and awe in national politics. A sane, once-a-day newsletter helping normal people make sense of the news.

I usually end up going here: whatthefuckjusthappenedtoday.com though @vox.com seems to have a new newsletter too.

I wrote this a month back, kinda knowing what was coming, hoping for the best, expecting the worst.

My cats are wondering why I have been extra cuddly lately 😅

It is such a slap in the face to the Indian American community to delay their green cards for decades and then declare that because of that delay their American children aren't citizens.

EB2 currently has an expected wait time of around 20+ years (around 24+ years after landing in US including H1B+filing time). Even EB1 for indians currently has 4ish years wait time (which you need a 5 year PhD to qualify for, and then an H1B, usually ending up being 10+ years since landing in US) 😅

Is it time for a social media break again? It has not been a great day. 😅

Yeah, I do like their rotis. I do not like their parathas or naans though. But they save so much time. Also, apparently vadilal makes great frozen roomali rotis? I just tried them out, and they are awesome!

Naan tacos? I want to start making more bread honestly, I am not used to it, so I end up eating just the salad, and then I crave chewy carbs 😅 (though haldiram's frozen roomali roti has been helping).