IAP AT GRIDSPACE 2024

Beyond LLMs: Is there anything they can't do?

So. Let's talk about LLMs.

Gridspace is running a month long intensive educational program on LLMs in industry. In this series, given by the engineers training, building, and designing products around these models, we will explore where we use LLMs, and where we don't. What makes them feel human-like, and whether it's an illusion. What we do to harness their strengths, and address their shortcomings. The series may also touch on the productionization of other models you've "seen in a Pset once", including in-house ASR, TTS, Emotion detection, Entity Recognition, Multi-Lingual models (...and more).

Who are we?

Gridspace is a product-based start up, born out of the eclectic landscape of LA. We are a small team of generalist and specialist engineers and designers, crafting production-standard speech and language technology to create commercial voice bots and speech analytics. This program is run by several MIT alumns now working for Gridspace, including post-doc physics students, and some good old course 6 undergrads and MEngs. Lectures will feature speakers from backgrounds spanning physics, linguistics, ML theory and dialog system design.

Where do I sign up?

Sign up using the following link!

Since our videos will be recorded and posted, we will need you to sign a basic waiver to allow us to post videos with you in them. We will be sending out a DocuSign in January, however you may also download the waiver from this link and return it, signed, to [email protected]

Logistics and Expectations

What can you expect of us?

• 2x weekly hour-long lecture talks at our offices in LA, which will also be streamed through YouTube. There will be the option to attend virtually, although we encourage anyone who is able to, to attend in person.
• Lecture topics covering everything from LLM tooling to philosophy to linguistics, given by MIT math postdocs to linguistics PhDs to Gridspace's cofounders and the engineers who have built these models.
• Open stage for questions, insight into building a voice bot, and to meet/chat with our engineers and designers
• An insight into the company ahead of our Summer Internship hiring cycle at Gridspace

What will we expect of you?

• Very little, beyond intellectual curiosity and enthusiasm.

2024 Schedule of Classes

Time: Monday^* and Thursday at 4:00pm to 5:00pm EST (Local Time: 1:00pm to 2:00pm PST)

^* Note that the first lecture of Week 2 will take place on Tuesday January 16^th, since Monday January 15^th is a holiday.

In-person Location: 1375 E 6th St UNIT 2, Los Angeles, CA 90021

IAP 2024 Lectures and Materials

Feedback Form

Week 1: Planning

Lecture 1: Can LLMs Plan?

Lecture 2: Tools for LLM Planning

Week 2: Memory

Lecture 3: Can LLMs Remember?

Lecture 4: Tools for LLM Memory

Week 3: Perception

Lecture 5: What do LLMs Perceive?

Lecture 6: Tools for LLM Perception

Week 4: Language & Symbolic Reasoning

Lecture 7: Philosophy of Generative Linguistics vs. LLMs

Lecture 8: Tools for LLM Conversations

Lecture 9: Can LLMs Do Math?

IAP 2023 Lectures and Materials

Project Files

Week 1: Fundamentals of Sound and Signals

Lecture 1: Sound, Acoustics and Voice

Lecture 2: Digital Signal Processing (DSP) and Information Theory

Lecture 3: Machine Learning and JAX

Week 2: Intro to Linguistics and NLP

Lecture 4: Morphology, Words and Tokenization

Lecture 5: Syntax, Semantics and Word Embeddings

Lecture 6: Large Language Models and InstructGPT

Week 3: Pillars of Speech Technology

Lecture 7: Automatic Speech Recognition (ASR)

Lecture 8: Text To Speech (TTS)

Lecture 9: Dialog Systems

Week 4: Real World Applications

Lecture 10: Emotion, Dialog Acts, Personality and Lying

Lecture 11: Conversation Design

Lecture 12: Deploying ML in the Real World

Company Information and Links