To understand why certain Dimensions and score level are chosen in a data annotation rubric for an LLM response evaluation, you need have an idea about taxonomies and its hierarchy and you'll understand how it can be used to indirectly inform those rubrics.
Most treat data annotation as just side gigs and fail consistently to obtain such. but I believe a proper portfolio of projects on data annotation and in-depth understanding of data annotation rubrics and benchmarks and why they are necessary and crucial during those interviews.
Im about to pursue something that has been due for 4 years , after my bachelors there were things that i wanted to do but couldnt do it , not beacuse i wasnt capable, but life didnt allow me to because life wanted to teach me few things before that ,now im ready for a new pursuit
@jimstr@Steve_Yegge Gatekeeping which the frontier labs definitely will !!! Cause they gotta maintain competitive edge also in the same of safety isnt it!!!
Any capabality of developing a frontier model by using LLM's that are developed by a closed source for profit frontier lab,will be restricked WHY?
1. To maintain the competitive edge so that no one else can build models as good as yours, using your own models.
2. Security, simple
As believers of open research, we are disappointed to see Anthropic silently degrading Fable 5 for AI development
"Any topic related to building pretraining pipelines, distributed training infrastructure, or ML accelerator design... may have limited effectiveness through Claude via methods such as prompt modification, steering vectors, or parameter-efficient fine-tuning."
Not only do they get to decide what you use LLMs for in research, but this also enables them to silently intervene in your research without you knowing.
This sets a dangerous precedent. If a model refuses openly, users can understand the boundary. If a model falls back to another model, users can still evaluate the difference. But if a model silently modifies or weakens its own answers while still pretending to help, researchers lose the ability to know whether a failed result came from their own idea, their implementation, or an invisible intervention by the model provider.
That is not safety. Safety policies should be transparent, auditable, and user-visible.
On top of that, the people most harmed by this are not the largest labs with massive teams and proprietary infrastructure. It is the independent researchers, academic groups, startups, and open-source builders who rely on public tools to compete, innovate, and pioneer AI for everyone else.
Frontier models out of Indian research labs is everything that needs to happen right now, if we need to see more foreign investors actually invest into indian Ai ecosystem, sadly that hasen't happned yet at the scale that needed to happen , but good to see labs are coming up.
#hiring for frontier AI research at @soketlabs
Compensation upto 1.5 cr
Access to thousands of GPUs
We are doing exactly what many told us not to do: building frontier AI out of India.
We're hiring exceptional researchers and engineers to help push the boundaries of foundation models in math, code, reasoning, and multilingual intelligence.
Open Roles
1. Research Scientist, Pretraining – Foundation Models: https://t.co/Aq4dH1Ac0B
2. Research Scientist, Data – Foundation Models: https://t.co/Vniq2Q0xop
3. Machine Learning Engineer – Foundation Models: https://t.co/PvBqB3j5dC
4. Kernel Engineer – Foundation Models: https://t.co/lMrmetParB
5. AI Data Engineer – Foundation Models: https://t.co/TSkSduWtbR
6. AI Data Curator – Foundation Models: https://t.co/xPksY4MQ3q
7. HPC Infrastructure Engineer: https://t.co/u5Uckr2tG2
#India #AI
Don't get scared when equations are thrown at screen , just note down and use chatgpt or gemini , preferably Gemini cause it literally teaches you how gradients flow and shows the math behind it step by step.
Just did a tiny variation while building a simple backprop ,used Sigmoid activation instead of tanh or ReLU ,now Andrej initially teaches it with tanhx but i felt sigmoid was easier to implement and understand.
you don't need a lot of calculus to understand backprop or Neural network and build intuition, my recommendation is just start with any video that teaches simple backprop (preferably andrej he helps you think and figure out things) and then google your way through missing stuff.
I've realised that if you try to learn backpropagation deeply to understand neuralnetworks , your brains will fire neurons in a way that they never could.
@levidiamode I've heard those assignments of cs336 are heavy would take months to complete is it ?? , I'm thinking myself of doing cs336 after finishing andrej karpathys playlist
@AHadzibabic Thinking of doing this after i finish Andrej karpathys playlist, but man cs336 assigments are HEAVYYY that should take few months to complete
Understanding neural networks from scratch ain't that hard , just try to spend more time understanding the mathematical intuition behind logistic regression and sigmoid function its pretty much easier after that , its just layers of the same function
One thing I've realised if you're trying to choose a career to get into in 2026 , don't go to youtube or anywhere and search is this a good career ? cause then a loop starts where you keep searching about the right career, but will never take the first step towards that career.
The amount of time I've seen posts regarding a new app that helps you find jobs, it's like one after the other and yet still people can't get jobs, it seems you can't solve this problem with any apps in 2026,
Either it's the candidates or the recruiting itself that's broken 🤷♂️