No images? Click here
Monthly AI Newsletter from PortNLP Lab
Issue #3 | January 2025
Welcome back to explain! the PortNLP Lab's monthly newsletter. As we step into the new year, we continue our commitment to advancing responsible artificial intelligence. This month, we highlight significant developments in the AI landscape and share exciting updates from our lab.
In the Know
đ° U.S. Policy Shifts on AI Regulation
In a notable policy reversal, President Donald Trump has rescinded the executive order on artificial intelligence risks previously established by the Biden administration. The original order mandated that AI developers share safety test results with the government prior to public release, aiming to mitigate potential risks to national security and public safety. The revocation has sparked a debate between promoting innovation and ensuring responsible AI development.
Read more
đ AI Open Source Advancement
AI lab DeepSeek has released its new R1 model family under an open MIT license, with its largest version containing 671 billion parameters. The company claims the model performs at levels comparable to OpenAI's o1 simulated reasoning model on several math and coding benchmarks.
Read more
đ˘ Sustainability Spotlight: AI's Carbon Footprint vs. Human Effort
AI training is highly energy-intensive, with models like GPT-3 emitting
over 500 metric tons of COâ, comparable to more than 100 gasoline-powered cars
over a year.
MIT technology review
However, once trained, AI can be significantly more energy-efficient than humans
for tasks like writing or illustrating. AI-generated content emits 130 to 1,500
times less COâ per page compared to human writing and 310 to 2,900 times less COâ
per image than traditional illustration.
Nature
While AI offers efficiency gains, balancing its energy demands during training with
sustainability efforts remains a challenge. Advancements in energy-efficient AI models
and reliance on renewable energy sources are key to minimizing its environmental impact.
Stay tuned for more insights on responsible AI practices!
Spotlight on Us
We're kicking off 2025 with some exciting research and community engagement! Hereâs what weâve been up to this January:
đŹ Fresh Off the Press
đ Cracking the code of Nepali idioms!
Can AI really understand the nuances of idioms in low-resource languages? We built neDIOM, a dataset that explores how AI interprets Nepali idiomsâsometimes successfully, sometimes hilariously.
âneDIOM: Dataset and Analysis of Nepali Idiomsâ by Rhitabrat Pokharel and Ameeta Agrawal made its debut at CHiPSAL, COLING 2025.
đ What really drives multilingual models?
We always hear âmore data is better,â but is that the whole story? Our latest work dives into what truly impacts multilingual model performanceâspoiler: itâs more than just data size!
Check out âBeyond Data Quantity: Key Factors Driving Performance in Multilingual Language Modelsâ by Sina Bagheri Nezhad, Ameeta Agrawal, and Rhitabrat Pokharel at LoResLM, COLING 2025.
đ Language Models quizzed with Linguistic Ambiguity
Speaking of âthe father of the baby who was driving to schoolââwho is behind the wheel? English syntax prefers âthe babyâ (though the little guy doesnât have his license yet!).
Itâs a battle of syntax vs. semantics in our poster âMissing the Cues: LLMsâ Insensitivity to Semantic Biases in RC Attachmentâ by Russell Scheinberg, So Young Lee, and Ameeta Agrawal at the Linguistic Society of America (LSA) in Jan 2025.
đ¤ Silicon Forest Tech Summit 2025
Prof. Agrawal delivered a lightning talk exploring the role of humans as AI evolves, reminding us that as AI continues to improve, it is the human touch that shapes the narrative. The talk emphasized the importance of responsible research and education.
Chief Editor of the Month: Sina Bagheri Nezhad
Contributors: Ameeta Agrawal, Ekata Mitra, Rhitabrat Pokharel, Russell Scheinberg, Yufei Tao
Follow us on
|