Thread regarding Fidelity Investments layoffs

It's actually pretty funny that we still call them 'Data Scientists.' Slapping 'scientist' in the title definitely strokes the ego, but let's be real,most of the work is just importing PyTorch and spinning up basic regression or classification models. A single prompt to Claude can generate 100x better code and models than what most of these guys build manually. We should probably just rename the role to something more accurate, like 'Model Integrators,' 'Data Chefs,' or 'Prompt Architects.

We have a sizable data science team, but they rarely build anything from scratch. Their actual work consists of downloading existing models, fine tuning them on our fidelity data, and patching together simplistic agentic workflows via the LLM gateway. To pad their metrics, they churn out pathetic whitepapers and low quality patents. The reality is that these are entry-level tasks; anyone with basic ML literacy could execute this workload without specialized expertise.

A Huge issue is that the industry is shifting. It values traditional DS expertise less and the ability to prompt an LLM in a system of code more. The trad DS folk have been mostly successful in brokerage in gate keeping the permission to prompt LLMs. They do their own one off inefficient evaluations of a prompt without allowing the engineers to help with low hanging fruit that could 10-100x accelerate the process via automated tools and pipelines. This cannot last because there are other BU's that do not have this territorial obstacle to process in place that will leap frog ( yes, geriatric fid frog can still leap on occasion) brokerage and the world outside regulated fintech is leapfrogging them, so thats being double leaped and for one I cannot wait to see them face plant in the dirt with excr-ment in their teeth to show for the wasted time and money.

Meanwhile, I will keep jekyl/hyde with Fidelity AI during the day and actual "the art of what is possible" by night until such time as I decide to take a 20-30% pay cut and do work that has actual value somewhere else.

BTW. WTF is the deal with people claiming we pay less? Is is just phone reps? For tech we pay very well, on par with almost/near FAANG companies, 30% higher than the RTP market. People compare inflated titles vs the tiels they could actually achieve at a 'real' tech co. Sorry, VP here, you are not a VP at Meta, but you can be a senior engineer at meta and be about the same comp.

@jq what have they done ? other than getting high salary and churning out junk papers ? Give 2-3 great papers or patents they have publish .If you cant then you also belong to that garbage

Data scientists make good coin. Get yourself educated and become one.

Most of them are churning our garbage in the name innovation .

If you’re not smart enough to understand their output you shouldn’t be the one judging their performance or value to the company. It’s the same thing over and over again with these posts. You’ve chosen a specific role because there is ONE person who is near you holds this title and you’ve decided their value is zero. I wonder what your answer would be if I asked you to describe this persons appearance?

In fidelity getting a patent= payrise + promotion, equates to $20k in more pay at the year end, hence people will do anything to get it here, stamp over others on the way, empowered by A0's and SVP's.

Real innovators are too busy actually building to waste time on the patent scam. Meanwhile, corporate "leaders" use the system to inflate their egos. They manufacture pathetic, loophole-ridden "inventions" just to farm clout and slap a fake innovation badge on their LinkedIn profiles. It isn’t progress; it’s a desperate, performative grift

The software patent system is a broken, fraudulent mess. It has devolved into a playground for legal maneuvering where companies we-ponize empty, jargon-heavy filings instead of real innovation. Because software patents are fundamentally unenforceable, they don't protect creators ,they just exist as worthless vanity trophies for corporations to collect. It’s time to scrap the entire program

Churning junk papers is a rational response to an irrational system, but it is a race to the bottom. The data science community is maturing; we are realizing that a paper is not a product launch. A true contribution should either shift our understanding of how models learn, or provide a rigorously verified, usable tool for practitioners. Until citation metrics are replaced by "reproducibility scores" or "impact-on-deployment" metrics, the churn will persist. In fact, the incentives that produce "junk data science papers" are directly descended from the incentives that produced the "software patent gold rush" of the late 1990s and 2000s.

Both systems suffer from the exact same pathology: measuring quantity over quality, combined with an examiner/reviewer bottleneck that cannot keep pace with the applicant's ability to generate combinatorial variations.
. Elite companies like Google, Amazon, and Microsoft realized 5–10 years ago that they were drowning in junk. They have shifted internal incentives:

Bonuses now trigger upon commercialization or successful litigation, not just filing.

They are aggressively publishing defensive publications (openly releasing the idea so nobody else can patent it) rather than wasting legal fees filing a patent for a feature they don't actually intend to use.

Promotion committees are starting to ask: "Was this patent ever implemented in a shipped product?" If the answer is no, the patent counts for zero.

But for the vast majority of mid-tier corporations, banks, and government contractors, the old game is still alive and well