For the 50th meme, here’s a little new classic 🧡

# Waste glass gets a second life in construction blocks | Researchers have developed a way to use powdered discarded glass in building blocks for construction, which could make this versatile material a lot more sustainable.

Is there a website or service that offers travel deals based on where you’re traveling FROM and not much else?

I have time off unexpectedly and thought about leaving town – don’t care if I go to California or Jamaica or New York or even Montreal – it depends on the price alone.

Am in Canada if this matters

I’ve been wrestling with a messy freeform text dataset using BERTopic for the past few weeks, and I’m to the point of crowdsourcing solutions.

The core issue is a pretty classic garbage-in, garbage-out situation: The input set consists of only 12.5k records of loosely structured, freeform comments, usually from internal company agents or reviewers. Around 40% of the records include copy/pasted questionnaires, which vary by department, and are inconsistenly pasted in the text field by the agent. The questionaires are prevalent enough, however, to strongly dominate the embedding space due to repeated word structures and identical phrasing.

This leads to severe collinearity, reinforcing patterns that aren’t semantically meaningful. BERTopic naturally treats these recurring forms as important features, which muddies topic resolution.

## Issues & Desired Outcomes

### Symptoms

* Extremely mixed topic signals.
* Number of topics per run ranges wildly (anywhere from 2 to 115).
* Approx. 50–60% of records are consistently flagged as outliers.

Topic signal coherance is issue #1; I feel like I’ll be able to explain the outliers if I can just get clearer, more consistant signals.

There is categorical data available, but it is inconsistently correct. The only way I can think to include this information during topic analysis is through concatenation, which just introduces it’s own set of problems (ironically related to what I’m trying to fix). The result is that emergent topics are subdued and noise gets added due to the inconsistency of correct entries.

### Things I’ve Tried

* Stopword tuning: Both manual and through vectorizer\_model. Minor improvements.
* “Breadcrumbing” cleanup: Identified boilerplate/questionnaire language by comparing nonsensical topic keywords to source records, then removed entire boilerplate statements (statements only; no single words removed).
* N-gram adjustment via CountVectorizer: No significant difference.
* Text normalization: Lowercasing and converting to simple ASCII to clean up formatting inconsistencies. Helped enforce stopwords and improved model performance in conjunction with breadcrumbing.
* Outlier reduction via BERTopic’s built-in method.
* Multiple embedding models: “all-mpnet-base-v2”, “all-MiniLM-L6-v2”, and some custom GPT embeddings.

### HDBSCAN Tuning

I attempted tuning HDBScan through two primary means.

1. Manual tuning via Topic Tuner – Tried a range of min\_cluster\_size and min\_samples combinations, using sparse, dense, and random search patterns. No stable or interpretable pattern emerged; results were all over the place.
2. Brute-force Monte Carlo – Ran simulations across a broad grid of HDBSCAN parameters, and measured number of topics and outlier counts. Confirmed that the distribution of topic outputs is highly multimodal. I was able to garner some expectations of topic and outliers counts out of this method, which at least told me what to expect on any given run.

### A Few Other Failures

* Attempted to stratify the data via department and model the subset, letting BERTopic omit the problem words beased on their prevalence – resultant sets were too small to model on.
* Attempted to segment the data via department and scrub out the messy freeform text, with the intent of re-combining and then modeling – this was unsuccessful as well.

## Next Steps?

At this point, I’m leaning toward preprocessing the entire dataset through an LLM before modeling, to summarize or at least normalize the input records and reduce variance. But I’m curious:

Is there anything else I could try before handing the problem off to an LLM?

EDIT – A SOLUTION:

We eventually got approval to move forward with an LLM pre-processing step, which worked very well. We used 4o-mini and instructed the prompt to gather only the facts and intent of each record. My colleague suggested to add the parameter (paraphrasing) “If any question answer pairs exist, include information from the answers to support your response,” which worked exceptionally well.

We wrote an evaluation prompt to help assess if any egregious factual errors existed across a random sample of 1k records – none were indicated. We then went through these by hand to verify, and none were found.

Of note: I believe this may be a strong case for the use of 4o-mini. We sampled the results in 4o with the same prompt and saw very little difference; given the nature of the prompt, I think this is very expected. The performance and cost were much lower with 4o-mini – an added bonus. We saw far more variation in the evaluation prompt between 4o and 4o-mini. 4o was more succinct and able to reason “no significant problems” more easily. This was helpful in the final evaluation, but for the full pipeline 4o-mini is a great fit for this usecase.

Hey all!

We’ve been working on an NLP tool for extracting argument structures (claims, premises, support/attack relationships) from long-form text like essays and articles. But hit a common wall: lack of clean, labeled data at scale.

So we built our own.

The dataset:

•1,500 persuasive essays

•Annotated with argument units: MajorClaim, Claim, Premise

•Includes labeled relations: supports / attacks

•JSON format with token-level alignment

•Created via an agent-based synthetic generation + QA pipeline

This is the first drop of what we’re calling DriftData and are looking for 10 folks who are into NLP / LLM fine-tuning / argument mining who want to test it, break it, or benchmark with it.

If that’s you, I’ll send over the full dataset in exchange for any feedback you’re willing to share.

DM me or comment below if interested.

Also curious:

• If you work in argument mining, how much value would you find in a corpus like this?

• Is synthetic data like this useful to you, or would you only trust human-labeled corpora?

Thanks in advance! Happy to share more about the pipeline too if there’s interest.

I love one bag travel, but unfortunately I have too many one bag hobbies that takes up their own bag or suitcase per hobby. Photography, scuba diving, etc.

If I want to leave things behind on long trips, short term luggage storage can get pricey. Security could be a concern with expensive equipment.

I had some success renting a long term storage for a SEA trip. I rented a monthly storage in Hong Kong and used HK as a hub. I reconfigure my packing depending on the trip. Scuba for Philippines, photo for Cambodia, one bag for urban destinations etc.

The degree of difficulties to rent monthly locker in HK is high. They require HKID, local address and bank account. Luckily I have the ID, can use relative’s address. They waived the banking requirement after dinner negotiation. But the deposit refund must go to a bank. So I gifted that to a relative. The storage outfit is one of many chains that operates in industrial parks. Have 24hr access and easily accessible via subway is key.

Overall, this is cheaper, safer and more convenient than short term luggage storage or airline baggage fees. Assuming the first and last leg are international flights with more reasonable baggage limit.

Do you have a similar travel hack? I would love to learn if there is a better approach with less degree of difficulties or cost. I think HK and Bangkok are my top options as travel hub for ease of entry and exit with a US passport.

Hi, I (23) am from India. I want to go into NLP/AI engineering however I do not have a CS background. I have done my B.A. (Hons) in English with specialised courses in Linguistics and I also have an M.A. in Linguistics with a dissertation/thesis. I am also currently doing a PG Diploma certifiction in Gen AI and Machine Learning.

I was wondering if this is enough to transition into the field (other than self-study). I wanted to go into research but I am not sure if I am eligible or will be selected in langtech programmes in universities abroad.

I am very confused about whether to get a job or pursue research. Top universities have fully funded PhD programmes, however their acceptance rate is not great either. I was also thinking of drafting and publishing one research paper in the following year to increase my chances for Fall 2026 intake.

I would like to state that, financially, my condition is not great. I am an orphan and currently receive a certain amount of pension but that will stop when I turn 25. So, I have a year and a half to decide and build my portfolio or CV either for a job or a PhD.

I am very concerned about my financial condition as well as my academic situation. Please give me some advice to help me out.