qi2peng2.bsky.social
Multimodal Agents Research @ Orby AI. Ex-AWS AI, JD AI. PhD from @stanfordnlp.bsky.social, UG Tsinghua U. He/him. Opinions my own.
53 posts
264 followers
42 following
Regular Contributor
Active Commenter
comment in response to
post
I was worried at first that the story required too much context form the MT original to resonate. It turned out a self-contained story with James' unique journey as he navigates racism and slavery, the voice acting brings the characters/personas to life esp you are not that familiar with AAVE.
comment in response to
post
Definitely! We have a job posting on our website where you can apply! Check out jobs.ashbyhq.com/orby-ai
comment in response to
post
solving them. As an RE, you'll work on state-of-the-art models & agents and infrastructure to support their training and serving.
If you have friends who might be interested, don't hesitate to forward them this information! Application links:
RS: https://buff.ly/4avCOm8
RE: https://buff.ly/4aC8qqo
comment in response to
post
with reasonable freedom, depending on the scale/focus of the business.
Case in point, we are looking to expand the research/foundation models team at Orby AI and are looking for highly motivated researchers and ML/Research engineers. Please reach out if you're interested in learning more!
/fin
comment in response to
post
the usual suspects, as well as many non-large-tech companies, including large companies not traditionally known for Internet/software products, AI-adjacent scale-up companies, and AI-native startup companies. Many of these can provide opportunities for exciting exploratory work
5/
comment in response to
post
were writing the research papers you are the most proud of. While many academic institutions are growing fast in recent years, there is likely still a fierce competition in more popular directions. In the meantime, there is actually a growing demand in the industry of AI talent, including
4/
comment in response to
post
2. Having more roles and players in the AI market does not mean fewer opportunities for PhDs, because the market is still in an upward trend. It does mean, however, that opportunities might be lying in places you hadn't thought about or anticipated when you entered the PhD or
3/
comment in response to
post
critical thinking, independence in problem solving, domain expertise, and vision about strategic technical directions are all valuable traits that takes a serious amount of training to obtain, and something one should be proud of and play to their own strength.
2/
comment in response to
post
1. Indeed, a lot of AI jobs in the industry will no longer require a PhD, because with scale and production comes processes and reliability requirements that do not require paradigm-shifting changes all the time. However, this does not mean a PhD training will make no difference --
1/
comment in response to
post
Here are the first two parts of the series in case you'd like to catch up:
- Part 0: Academia vs Industry https://buff.ly/3W6S0A9
- Part 1: How to choose a team https://buff.ly/3W4mFy4
/fin
comment in response to
post
publishing (or more specifically, the lack thereof) in the industry, what are the implications for individuals, and what the day-to-day might look like for industry researchers. Hope this provides useful perspective: https://buff.ly/400ITlG
3/
comment in response to
post
... in this series, which actually aims to answer the original question in the series title: "What does an industry researcher do, anyway?"
In this post, I attempt to answer some questions my younger self had about
2/
comment in response to
post
-- these will likely compound and pay dividends down the road.
(For instance, I learned a lot about bash scripting from some early collaborations at Stanford, and experimental work has never been the same if I only knew how to do everything in, say, Python.)
/fin
comment in response to
post
If you are yearning for interesting new challenges down the road, never settle for complacency -- look for things in your current work to expand your exposure to problems and how they are solved, even if it's a small step that makes your current work a little easier
3/
comment in response to
post
It is often through experiencing new tasks and challenges either first-hand or as witness that we acquire new skills, new mindsets, new ideas, or even just new heuristic rules-of-thumb for them, which we add to our toolkit and eventually benefit from in solving new challenges. 2/
comment in response to
post
who was making extrapolations from known scientific models and claiming such categorically unobservable extrapolations must be reality.
Hope you'd like it, and Happy Holidays!
comment in response to
post
"science won't be able to explain everything and answer every question", but just often not in the way that people constantly claim this might think. Also, it helps offer a clear debunk of some of the very ascientific claims that another prominent physicist made in one of their books,
5/
comment in response to
post
One topic that strongly resonated with me is how Dr. Hossenfelder makes the clear distinction between "scientific" and "ascientific". A lot of theories that are, by definition (of science), ascientific, so one would be correct to claim that
4/
comment in response to
post
This book really resonates with me, and explains a lot of the concepts I feel the urge to yell out when I see pseudo-science and exaggerated science reporting in mainstream media -- only Dr. Hossenfelder did it a lot more eloquently already.
3/
comment in response to
post
and explaining what "science" is and does in the meantime. I'd strongly recommend it as a good brush-up for practicing scientists, or those around you that might be curious about either the philosophical side of it or the scientific side.
2/
comment in response to
post
Dr. Sabine Hossenfelder's ๐๐ญ๐๐จ๐ฉ๐๐ฃ๐ฉ๐๐๐ก ๐๐๐ฎ๐จ๐๐๐จ: ๐ผ ๐๐๐๐๐ฃ๐ฉ๐๐จ๐ฉ'๐จ ๐๐ช๐๐๐ ๐ฉ๐ค ๐๐๐๐'๐จ ๐ฝ๐๐๐๐๐จ๐ฉ ๐๐ช๐๐จ๐ฉ๐๐ค๐ฃ๐จ is a very approachable popular science book talking about some of the questions at the heart of philosophy since the dawn of philosophy,
1/
comment in response to
post
or related fields. You should have a strong background in machine learning and deep learning and hands-on experience implementing these algorithms. Experience in NLP, computer vision, multimodal systems, and agentic systems, or a strong track record of publications is a plus.
comment in response to
post
or exciting applied research achievements. Please send your resume to [email protected] if interested!
The internship will be based in Mountain View, CA, USA. You must be enrolled in a PhD or Masterโs degree program in Computer Science, Machine Learning, Operations Research, Statistics,
2/n
comment in response to
post
What happens if some crucial Cloud or AI capabilities you rely on goes down, and how do you limit the downstream impact to your customers? What can we learn from infra providers like utility companies for what (not) to do in building&maintaining infra, or responding to incidents?#FoodForThought
/fin
comment in response to
post
As enterprises, especially SaaS companies, I believe there is also much we can learn and reflect from such incidents. SaaS companies are often both consumers and providers, with not only #compute & #storage, but more and more #AI being part of our web infrastructure.
5/n
comment in response to
post
how to we make sure these infrastructural systems more robust and redundant to ensure uptime; and as users, how can we be reasonably prepared for disruptions without going all out to building billionaire doomsday bunkers?
4/n
comment in response to
post
These are quite challenging numbers to uphold, and failure to upholding them can result in severe worst-case outcomes. This is by no means a call for every modern person to learn Paleolithic ways to survive, unless that's your calling or hobby. I'm more curious -- as a modern society,
3/n
comment in response to
post
Electricity, tap water, waste water plumbing, home-use fossil fuels -- these are just a few things we have come to depend on reliably. How reliable are they? Take typical numbers in modern software Service Level Agreements (SLAs):
95% uptime = 18 days down/yr
99% = 3 1/2 days
99.9% = 9 hours
2/n
comment in response to
post
This shows strong performance gains over simple web agents on a number of benchmarks, and improves the efficiency of inference for the agents significantly. See Yu Gu's X post for more details (and the list of our awesome collaborators): https://buff.ly/411y2tR
/fin
comment in response to
post
a world model of the Internet, where it helps the agent simulate imagined world states after it performs proposed actions in the environment, which helps better understand the effects and rewards each candidate action can potentially derive from the current state of the world.
4/n
comment in response to
post
In Orby AI's latest collaboration with the OSU NLP group, ๐๐๐๐๐ซ๐๐๐ฆ๐๐ซ, we tackle this problem by leveraging #FoundationModels as world models for inference-time search. Specifically, we find that via clever prompting and task formulation, we can turn frontier models into
3/n
comment in response to
post
This is not always trivial, because for GUI tasks, accurate simulation of the environment is often difficult due to the complexity of the applications, and inference-time search infeasible since many actions have irreversible side effects (e.g., placing an order).
2/n
comment in response to
post
from very different angles/backgrounds. If you are someone with diverse backgrounds, this might be a good time to revisit some of your roots!
NOT part of an incumbent team? Don't worry! There're always interesting research problems-- just don't be tricked to look in the same direction they are!
/fin
comment in response to
post
4. There will always be smart improvements coming from different angles to solve existing problems much better in a (performance, efficiency, data cost, ...) combination in a David vs Goliath manner rather than incrementally improving upon them, which usually requires looking at the same problem
6/n