Surge AI Blog

I'm a PM at a human data company (https://www.surgehq.ai) that helps the large language model companies ensure their models are safe (we're the “clever prompt engineers” who helped Redwood assess their model performance). We actually just published a blog today that includes our perspective on building “AI red teams” and best practices for AI alignment/safety: https://www.surgehq.ai/blog/ai-red-teams-for-adversarial-training-maki…

I'd love to chat. Want to reach out to the email in my profile? I'm the founder of a startup solving this exact problem (https://www.surgehq.ai), and previously built the human computation platforms at a couple FAANGs (precisely because this was a huge issue I always faced internally). We work with a lot of the top AI/NLP companies and research labs, and do both the "typical" data labeling work (sentiment analysis, text categorization, etc), but also a lot more advanced stuff (e.g., search evaluation, training the new wave of large language models, adversarial labeling, etc -- so not just distinguishing cats and dogs, but rather making full use of the power of the human mind!).

Surge AI (https://www.surgehq.ai) | Full-Stack | Remote or San Francisco / New York Love language? So do we, and our mission is to infuse AI with that same love. At Surge, we're building the human infrastructure to power NLP — from detecting hate speech, to parsing complex documents, to injecting human values into the next wave of language models. Our first product is a platform that helps ML teams create amazing, human-powered datasets to train AI in the richness of language. We're a team of former Google, Facebook, and Airbnb engineering leads, and we work with top companies at the forefront of machine learning. Our tech stack is Ruby on Rails, React, and Python. We’re rapidly growing, and we're looking for full-stack engineers to join the team and develop our product. To apply, please email [email protected] with a resume and 2-3 sentences describing your interest in Surge. We love personal projects and writings too! More information: https://www.surgehq.ai/about#careers A blog post explaining the problems we are working to solve: https://www.surgehq.ai/blog/the-ai-bottleneck-high-quality-human-power…

AI writing you can recognize as AI writing is obvious. Newer models are better about this and the line will only get more blurry. Here's a benchmark where good writers make the assessment rather than different LLMs ranking each other: https://surgehq.ai/leaderboards/hemingway-bench The top models are also the latest: Gemini 3.1 Pro: still a bit of a gremlin, but will probably stay on top until the other model makers go xkcd 810 and target this benchmark Gemini 3 Flash: current favorite of writers using it as a helper for its speed and decent prompt following

Hmm, I don't want to talk specifics about my experience, but maybe check out some of the case studies on Surge's website - https://www.surgehq.ai/ (about halfway down the page). > Do they ask PhD to explain root of negative 1 and why is it complex? This isn't necessarily impossible, but I would consider it to be infeasible with existing labelling workforces. Of course if you really needed a dataset like this and you were sufficiently resourced and willing to spend, you could maybe make it work (I would question whether you really needed PhDs though, that might be hard to swing at any price point). But the core idea behind your question is correct - this is what a dataset might look like and hiring/contracting appropriately-skilled people and asking them to do repetitive tasks with some guidance is how you would go about getting it. Depending on the need, it can be quite a bit more complex too - if you needed self-driving car driving behavior data maybe you build a simulator and hire people to drive in the simulator and use that as training data (made up and probably crap example, but it illustrates the possibilities). Some people think that labelling workforces are all low skill and there is a lot of good things low skill workforces can do well (visual stuff, basic language and emotion tasks), but you might be surprised at the ability to get skilled labelers. There are lots of smart/educated people around the world and there is ridiculous amounts of money flowing into…

Surge AI (https://surgehq.ai) | Software Engineers (full-stack or applied ML) | Full-time | SF or Remote Our mission at Surge is to build the human infrastructure powering the next wave of AI and LLMs. We’re building a data platform that powers AI teams at OpenAI, Anthropic, Meta, Google, and more. Reinforcement Learning with Human Feedback is the critical technique behind the new generation of AI assistants, and that human feedback comes from us. Our product has been a game-changer for the top AI teams in the world. Here are some examples of our past work: - Creating the GSM8K dataset: https://www.surgehq.ai/blog/how-we-built-it-openais-gsm8k-dataset-of-8… - Collecting data for Anthropic's Claude: https://www.surgehq.ai/case-studies/anthropic-claude-surgeai-rlhf-plat… You’d be joining a small, rapidly growing team of former engineering and ML leaders from Google, Meta, Twitter, and Airbnb. We ship quickly, deeply care about the problems we are trying to solve, and value autonomy and ownership. No previous AI experience is required, if you have the engineering skills you can learn what you need on the job. We're looking for engineers with a few years of experience and you can work out of our office in SF or remotely. Please reach out to us at [email protected] with a resume and 2-3 sentences describing your interest. Excited to hear from you!

Frontier models need to do everything for everyone. It's expected (though not often done) that smaller models fine-tuned on specific tasks can approach frontier performance on a specific area. [0] Post-training/fine-tuning is not trivial and having it as a service might make it more accessible. [0] https://surgehq.ai/blog/training-on-complexconstraints

link broken (extra slash) https://www.surgehq.ai/blog/the-pdf-that-broke-chatgpt

I thought Claude used human feedback due to Surge claiming they were a customer: https://www.surgehq.ai/case-studies/anthropic-claude-surgeai-rlhf-plat…

Mixtral is missing in half of the benchmarks in that paper. Hardly conclusive. It’s also common knowledge that these benchmarks have a lot of issues[0]. A good litmus test, but not a substitute for actually seeing how the models do in the real world. On the topic of “hardly conclusive” things, Gemini Pro literally told me just a few minutes ago[1] that the Avatar movies did not have humans in them. There was no funny business in the prompting. At least Mixtral knows that Avatar has humans in it. Most of Gemini Pro’s responses have been fine, but not exceptional. [0]: one random article talking about these issues: https://www.surgehq.ai//blog/hellaswag-or-hellabad-36-of-this-popular-… [1]: https://i.imgur.com/En37EJD.png

These systems typically have major issues with bias and mislabeled training data [0], and I have a feeling ICE has very lax heuristics for what makes a given post extremist or indicative of criminal activity. My own views are far from moderate in many respects, and I don't think ICE would like them very much. [0] https://www.surgehq.ai//blog/30-percent-of-googles-reddit-emotions-dat…

Hellaswag is also a deeply flawed benchmark, I wouldn't read too much into it: https://www.surgehq.ai/blog/hellaswag-or-hellabad-36-of-this-popular-l…

Likely. It appears to be part of the company's potential toolbox and has been researched by them for their blog: https://www.surgehq.ai/blog/humans-vs-dall-e

I quite like https://surgehq.ai/benchmarks/hemingway-bench