Key: (M)onday (T)uesday (W)ednesday Thu(R)sday (F)riday (S)aturday S(U)nday Also note: order of log is reverse chronological order, newest first.

Note: The above snippet is stolen from my advisor’s 2nd year PhD progress log 🤣

April

Week 13

Weekly summary:

Research:

• Systematically studied LLM for the first time.
  • Outcome: a notion page summarizing it. Internal to WING. Upon request.
• Finally have had my “discourse day”.
• Proposal for my parallel project
  • [ ]
• Dataset section for ELCo . Simultaneously got following done:
  • Forward / backward thinking. Forward: How will people text in 5 years later? Backward: What is the reason for our COLING rejection. How can I revise? What lessons can Ziqing learn from that?
  • Corpus study [ ]

Misc:

• Finished my teaching duties this semester.
  • Gave final tutorials to three TGs.
  • Busy Monday helping my mentored teams revising their final reports. Some are in a pretty good shape.
• Bugged (literally 🐛) by bugs in Botanic Garden last week. Went to UHC twice.

Week 12

Forgot to record.

March

Week 11

Weekly objective (O):

• O1: Get ELCo paper draft ready to submit (to arXiv / any suitable conference). Present to Min next Monday.
• O2: Think about my ACL-23 rejection and improvements/generalization (maybe set aside one day only doing this).
• O3: A concrete plan for my next work. Propose it to Min next Monday.
• O4: CS4248 TA responsibilities:
  • A2 grades release (myself as the primary grader).
  • A1 rebuttal reflected in Canvas.
  • Teach tutorials and mentor project teams.
  • Design exam questions.

Weekly review:

• O1: [50] /100
• O2: [50] /100
• O3: [50] /100
• O4: [100] /100

03-28 M:

Week 10:

03-27 U:

• Went to Liangming’s Lunch organized by Yajing.
• Went to Botanic Garden to enjoy Gin and Tonic made by Zejun (takeaway package!).

03-26 S:

• Submitted my author feedback to ACL-23.
• Went to 歌谣 Geyao 2023 where featured Longshen 🎻. It was pretty nice.
• A playlist for my recording: https://www.youtube.com/playlist?list=PLNsYEMHIB4shDAL20_Q4VydWyOZ2JID0-. Geyao people have nicely agreed this playlist to go public (previously unlisted).

03-22 ~ 03-25 W to F: Forgot to record on a daily basis …

• Tried to make ELCo to *SEM but didn’t make it 🥲

• Friday’s group meeting: Yiran’s talk is closely related to me. Think about how to improve? Arranged a discussion on 31st March.

03-21 T:

• Still think about Min’s comments.
• Try out Midjourney (OK … I spent too much time on it. Need have a hard constraint of the time!)
  • OK … Even though I wasted some time, I have a few takeaways!
  • Midjourney is very good at generating open ended prompts, like, “Two PhD students (male, asian) are discussing Midjourney in their lab (CS department).” Image here: https://cdn.discordapp.com/attachments/1087630864789295155/1087969374066970686/yisong_Two_PhD_students_male_asian_are_discussing_Midjourney_in_ca29cd94-3b56-48c7-93e7-e39ce536c974.png
  • But it fails at prompts requiring a latent structure, for example:
    • A young man (black hair) is drawing on his iPad. What he is drawing is a chocolate cake. The result ends up like a man is eating the cake on top of an iPad with Apple pencil (as his fork). Image here: https://cdn.discordapp.com/attachments/1087630864789295155/1087710774308376687/yisong_A_young_man_black_hair_is_drawing_on_his_iPad._What_he_i_bf7e688f-dc13-4063-985e-377fa68edd65.png
    • A beautiful girl is drawing a profile photo for her boy friend. The result ends up like a girl is drawing a picture for another girl.
  • ⭐️ It’s interesting to think about how to incorporate such latent structures into the prompts and inform the model ( ⭐️Min also told me this is an important problem. )

03-20 M:

• Met Min, review my ACL-23 review.
• Think about Min’s comments.
• Talk to Jinlan.

03-15 W:

• TG05, 06, 07, 08, 09. (a very tough 5 hour session, hopefully I can get some help from Lin).

03-14 T:

• George’s seminar.
• TG02.
• ELCo revision’s sprint w/ Ziyun.

03-13 M:

• Met Min.

03-06 ~ 03-10 (Week 8):

• NSSDM post-production.
• ELCo.

02-27 ~ 0303 (Week 7):

• NSSDM organization.
• Assignment 1 grading.

Feburary

02-24 F:

• Release ELCo to project teams. It turned out that 7 teams are interested. Ziyun and I have created something interesting to people.
• Group meeting.

02-23 R:

• Reflect on the meeting w/ Min. Stratify my research into 3 layers.

02-22 W:

• ⭐️ Zoom w/ Neo about GPTs and ICL. Potential collaboration w/ OpenAI in the future.
• Video w/ Yahui, for queries about my undergrad institute and discuss her current study of Mooc.
• Kick off my project mentoring meeting with all 5 teams I am working with (A very tough 4-hour session. I didn’t sleep well …). I like some of the teams’ initiatives. Some of them already have their ideas. Some of them come up with their own ideas after a few “prompts” from me. Some of them need more help.

02-21 T:

• TA meeting.

02-20 M:

• ⭐️ Met Min.
• Dinner w/ Qizhe.

02-19 U:

• Prepare for tomorrow’s meeting w/ Min.
• Emailing to NLP project teams I mentored. Asked Ziyun’s permission to share her FYP defense slides as a guide for students studying her dataset.

02-17 F:

• ⭐️ Met Q.
• Read Q’s papers.

02-16 R:

• Present at George’s seminar about ChatGPT.
• Audiences (most are entrepreneurs) liked my presentation a lot, especially the theory of LMs and how I explained why ChatGPT is stronger than previous LMs.
• Their burning question: How to specialize ChatGPT for their own business?
  • My answers: it really depends on how OpenAI plans to opensource ChatGPT. (1) Decoding strategies (if NOT opensourced); (2) Domain-specific fine-tuning (if opensourced).
• I highlighted the ethical concern and responsible use of ChatGPT.

02-15 W:

• Reply to students’ email/TG messages for my TA class.
• Prepare for a meeting w/ J. Met J and discuss about my research.
• Prepare for my presentation tomorrow.

02-14 T:

• J’s visit. Seminar, lunch and PhD meetings.

02-13 M:

• Read papers.

02-11 S:

• Meetup w/ Qian and Longxu in Botanic Gardens. Recieved a lot of useful feedback!

02-10 F:

• Present and pilot my annotation task in group.

02-09 R:

• Prepare for my annotation task.

02-08 W:

• TG05/06.
  • It is the first time Min sits in my tutorial. I felt a bit nervous. But it completed OK.
  • Min gave me feedback for my teaching skill (I need to save them if I want to be Prof later …)
• Preparation for my annotation task in group meeting.
  • Design done. (Slides skeleton, annotation workflow).
  • Not clear about the definition of simulatability. It is to simulate the ground truth answer, or to simulate model’s prediction (which is not necessarily true)?
• Continue reading thesis.

02-07 T:

• TA meeting and TG02.
• Read papers by Dan Iter and Karthik Narasimhan.

02-06 M:

• Meet Min w/ Longxu, Tongyao, and HC.
• I received a ton of feedback (an intense discussion! don’t know how Min managed that everyday).
• Summarized the meeting in QA style. Check Meeting 6 Feb 2023 in my personal notion.

02-05 U:

Prepare for tomorrow’s meeting w/ Min:

• Read my 1-man slack’s message about all ideas / problems I recorded for discourse.
• Chat w/ HC.
• Slides done. Come up one core question to ask Min tomorrow with evidence supporting that.

02-03 F:

• Attend Longxu’s meeting.

02-02 R:

• Read papers esp by Qian.
• Longxu and Qian invite me to collaborate on my research!

02-01 W:

• Thinking about a fine-grained experiment section for ELCo. 50% done. I find Qing’s paper @ NAACL’22 a very good example from meta-level.
• Study Chuyuan’s paper, the idea of attention is interesting.
• Reply to students’ email.

January

01-31 T:

• TA meeting (I find the traffic is much faster in early morning).
• Give T01-T04.
• Help Tutu’s submission a bit.

01-30 M:

• Prepare for the tutorial.
• Read a few *SEM papers.

01-18 W:

• Showed my manuscript to Liangming. Received many insightful comments!

  Takeaways:

  • On the position of this manuscript (it’s still a bit unlcear, but there’s little I can do before ACL deadline).
  • The idea is cool and valid.
  • On my long-term tasks. Think about involving more challenging reasoning, creating a new task with more challenging examples. Otherwise it’s less space for our “ordinary NLPer” to do in the era of GPT3/ChatGPT.

• Think about Liangming’s feedback. Read the EMNLP ‘22 paper, I find it’s an easy idea with straightforward execution.

• Try to revise some minor points …

01-17 T:

• Paper revision.
• TA meetings.

01-16 M:

• Paper revision.

01-15 U:

• Paper revision.
  • Try to change the discourse structure of the paper.
  • Abs + Intro 1st interation of the revision.

01-14 S:

• Family day. Went to my Grandma’s house. Met my uncle antie and my cousin with my parents.
• I think my driving skill has upgraded to a new level. I drove all the way smoothly.

01-13 F:

• Group meeting. Chenglei gave a very nice presentation. I learned a lot from it.

01-12 R:

• Read papers.

01-11 W:

• Prepare for meeting w/ Nancy and Wenqiang.
• Meet Nancy. Receive encouraging feedbacks.

01-10 T:

• TA meeting for CS4248.
• Read Yoav’s note on LLM: https://gist.github.com/yoavg/59d174608e92e845c8994ac2e234c8a9

01-09 M:

• Prepare for the meeting w/ Min.
• Meet Min.

Major takeaway from the meeting:

• Min told me it’s better to be explicit (even though reviewers might disagree), than to be ambiguous and confuse the reviewer.
• I asked Min so how big a dataset shoud be? Min told me the size can be justfied by the usefulness (e.g. improvement on another task).

01-08 U:

• Prepare for the meeting w/ Min.

  Major progress: Compare my work with two recent papers and think of a new argument. Come up a new paper map for Zi Yun’s rejected manuscript.