TeamBlind GPT Analysis

Community Development
1

Based on a comment from @Anonyman earlier about how rumors “leak” on blind I wanted to take a swing at scraping results from the site and forwarding them to GPT for analysis. The results were extremely promising:

Companies:
Microsoft: Referenced 6 times. Summary: Comments discuss layoffs at Microsoft, the severance package offered, and the impact of layoffs on employees.
Amazon: Referenced 5 times. Summary: Comments discuss layoffs at Amazon, the severance package offered, and the impact of layoffs on employees.
Google: Referenced 5 times. Summary: Comments discuss layoffs at Google, the severance package offered, and the impact of layoffs on employees.
LinkedIn: Referenced 1 time. Summary: Comments discuss rumors of layoffs at LinkedIn.
DoorDash: Referenced 2 times. Summary: Comments discuss layoffs at DoorDash, the impact of layoffs on employees, and the difficulty of finding a new job.
Rippling: Referenced 1 time. Summary: Comments discuss rumors of layoffs at Rippling.
Merck: Referenced 1 time. Summary: Comments discuss layoffs at Merck.
PVH: Referenced 1 time. Summary: Comments discuss the impact of layoffs on employees.
Amplitude: Referenced 1 time. Summary: Comments discuss layoffs at Amplitude.
Deloitte: Referenced 1 time. Summary: Comments discuss layoffs at Deloitte.

The above picked out that rumors of layoffs at LinkedIn and Rippling were being discussed and was able to pick out how often each company was mentioned. This is sort of a proof of concept example for other things we’re working on too such as EDGAR API Scraping/Alerting Ideas.

I’m unfamiliar with blind but I know many of you probably look through it somewhat often. So the question is what types of things should we push this process towards looking for?

5 Likes
2

another idea similar to this would be for earnings. I see tf talking about MSFT right now so my though was a day before, week before, whatever amount of time really, scrape blind or any other relevant sites for earnings keywords. Not sure what/which data you are feeding gpt but dumbed down example would be if your looking for “layoffs” instead look for “earnings” as well as whichever companies are on the calendar for the given week.

1 Like
3

obviously theres rumors on social media and earnings whisperer type stuff but might be worth it to try a quick analysis on whatever is ah today just to see if anything was meaningful

4

I’ve expanded this function to take into account the trending page as well and will probably have it all the available categories for their topic trending posts and then we can run further analysis on the data collected in multiple contexts if need be. The current iteration of this tool scrapes the comments → breaks them out into company → conducts analysis on the comments of each company all together. From this it generates a set of data:

  1. Company Name
  2. A count of the comments
  3. An analysis of whether the overall sentiment of the comments is bullish or bearish
  4. A predicted % move in the stocks price based off of the analysis
  5. A summary of the comments and analysis

The results of the first scan on 58 data points is below:

Netflix | 1 Comments | Bullish | 2-3% Move
The comment indicates that the company is supportive of employees with children, which could be seen as a positive sign for the company’s culture and could lead to a slight increase in stock prices.

Snap | 1 Comments | Bearish | -5% Move
The comment indicates that the company is considering layoffs, which could lead to a decrease in stock prices. Historically, news of layoffs has caused stocks to drop by an average of 5%.

Microsoft | 5 Comments | Bearish | -2-3% Move
The comments indicate a bearish sentiment for Microsoft due to the news of 80K out of 200K layoffs being people on work visas. This news could lead to a decrease in stock prices of 2-3%.

Salesforce | 2 Comments | Bearish | -2% Move
The comments indicate that Salesforce has overhired and is now laying off employees, which could lead to a bearish move in the company’s stock. Historically, news of layoffs has caused stocks to drop by an average of 2%.

Booking.com | 1 Comments | Bullish | 2-3% Move
The comments indicate that the company is providing a good work environment with generous benefits, which could lead to a positive sentiment from investors and a potential increase in stock prices.

Intuit | 1 Comments | Bearish | -2% Move
The comment indicates that Intuit is having difficulty managing their YouTube ads, which could lead to a decrease in revenue and a bearish move in the stock. Historically, news of this nature has caused stocks to move down by an average of 2%.

Meta | 2 Comments | Bearish | -2% Move
The comments indicate a bearish sentiment towards Meta, as the employees are expressing feelings of hopelessness and difficulty in finding a job. Historically, news of layoffs and difficulty in finding employment has caused stocks to drop by an average of 2%.

New | 5 Comments | Bullish | 2-3% Move
The comments indicate that the company is well-liked by its employees and is seen as a supportive employer. This could lead to a positive sentiment in the market and a potential increase in the company’s stock price of 2-3%.

Apple | 5 Comments | Bearish | -5% Move
The comments indicate a bearish sentiment towards Apple, as they reference layoffs, exploitation of the little guy, and human trafficking. Historically, news of this nature has caused stocks to drop by an average of 5%.

LinkedIn | 2 Comments | Bearish | -2% Move
The comments indicate a bearish sentiment towards the company, as they reference layoffs and a negative opinion of the CEO. Historically, news of layoffs and negative sentiment towards a company’s leadership can lead to a decrease in stock prices of up to 2%.

MasterCard | 1 Comments | Bullish | 2-3% Move
The comment indicates that MasterCard is being used to book travel, which could indicate increased consumer spending and confidence in the company. This could lead to a bullish move in the company’s stock of 2-3%. Historically, news of increased consumer spending has been associated with a positive move in stock prices.

Amazon | 10 Comments | Bearish | -2-3% Move
The comments indicate a bearish sentiment towards Amazon, as many of the comments reference layoffs, job market uncertainty, and the lack of loyalty from employers. This could lead to a decrease in stock prices of 2-3%, as news of layoffs and job market uncertainty have historically caused stocks to drop.

AbbVie | 1 Comments | Bullish | 2-3% Move
The comment indicates that the employee is making more money than their grandparents, which could be seen as a sign of the company’s success and a potential increase in stock value. Historically, news of a company’s success has been known to increase stock prices by 2-3%.

Reddit | 1 Comments | Bearish | -2% Move
The comments indicate that the company has recently started layoffs despite previously claiming that there would be no layoffs. This could lead to a bearish move in the company’s stock as investors may be concerned about the company’s ability to manage its workforce and its financials. Historically, news of layoffs has caused stocks to drop by an average of 2%.

Google | 5 Comments | Bullish | 2-3% Move
The comments indicate a positive sentiment towards Google, with references to the company’s success and its employees’ loyalty. This could indicate a coming bullish move in the company’s stock, with a potential increase of 2-3%. Historically, news of this nature has been known to move stocks in a positive direction.

SAP | 1 Comments | Bearish | -2% Move
The comment indicates that employees are not in favor of unions, which could be seen as a sign of discontent among the workforce. This could lead to a decrease in productivity and morale, which could lead to a bearish move in the company’s stock of around 2%. Historically, news of employee discontent has been known to lead to bearish moves in stocks.

Chainlink Labs | 1 Comments | Bearish | -2% Move
The comment indicates that the company is not doing well, as it references layoffs and a lack of sympathy for the employees. This could lead to a bearish move in the company’s stock, as investors may be wary of investing in a company that is not doing well.

Expedia Group | 1 Comments | Bullish | 2-3% Move
The comment indicates that the company is doing well in the current market, which could lead to a bullish move in the stock. Historically, news of a company doing well in the market has led to a 2-3% increase in stock prices.

Affirm | 1 Comments | Bullish | 2-3% Move
The comment indicates that people are looking to buy homes and are likely to use Affirm’s services to do so. This could lead to an increase in revenue for the company, which could lead to a bullish move in the stock price of 2-3%.

Going to require some further optimization but I think this is a promising start. An addition I think is to have things further categorized, one of the categories being “rumor”. The above data is returned as JSON so Mimir can act on it in varying ways. The system is also built so that it can include data from additional sources with ease.

While potentially optimistic, I think the end goal for this is going to be a few key points. Firstly, having Mimir actively scan for rumor so that signals may be generated about potential news. Secondly and I think the more optimistic scenario but I’m confident it can be done from what I’ve seen so far, is to keep an evolving summary of the current state of a company’s sentiment so that we can track sharp changes over time.

Going back to the above data, one of the key points is that it assigned arguably the most material item it scanned the largest percentage move estimate which was SNAP at -5% on the discussion of layoffs. In the current market the sentiment it assigned is wrong but there isn’t an “environment prompt” assigned to this task as of yet. Including one would easily fix that issue.

This project has given me a lot of hope for the SEC filing stuff we’re looking at doing.

Let me know if anyone has any thoughts.

1 Like
5

I finished the scraper so it now scans all pertinent links on Blind. This kicks the comments up from 58 to 606. Using the newly scraped set of data, I created a oneshot analysis to scan for rumors, the results of which are below:

Sambanova: Rumors of a fire sale and board looking for a bag holder, which could lead to investors not recouping their money and employees with common stock getting nothing.

NVIDIA: Rumors of a Google hardware layoff and its impact on the hardware division and Pixel teams.

Qualcomm: Rumors of HR ghosting in the middle of negotiations and hiring being frozen.

Lockheed Martin: Rumors of employees not being able to get clearance applications approved and the potential of being let go if denied.

Yahoo: Rumors of Google potentially laying off 30000 in the next 6 months.

Blizzard: Rumors of potential layoffs.

343 and Bethesda: Rumors of layoffs despite the company being understaffed and Starfield coming out soon.

Southwest: Rumors of bad management and systems.

SpaceX: Rumors of unprofessional behavior after a final interview.

Cummings Group: Rumors of a hire to fire offer and potential layoffs.

Going to need to build in some matching to ensure we’re dealing with listed companies but I’m pretty happy with the results thus far.

3 Likes
6

Did a bit of a rework on a part of this system today to get the cost per run down. It used just over 150K tokens yesterday for a cost of ~$2.50, todays run dropped it to 12K tokens and ~$0.25.

Out of 698 comments captured, GPT was able to sort through and identify one instance of a comment containing information that could materially effect the price of the stock:

image

This was captured from this comment which linked a news article from a couple days ago (the comment is older, there are no date filters at the moment because I’m looking for a wide set of test data):

link
Unity Layoff
Unity to layoff 300 employees - probably HR and operations. Unity Software to lay off nearly 300 employees #unity New pizzzzzzza 670 6 5 Jan 17 Bookmark

I think the criteria is a little too tight at the moment, but I remain solidly impressed by the functionality that is being brought to the table. This system is now live and will be running daily.

I’m currently treating the rumor harvesting as a separate task from the “summarizing” function I showed previously. The summarizing function is still under development as I’m not sure how I exactly want to utilize it as of yet.

4 Likes
7

This is interesting to some degree for sure. Layoffs are gonna be the bulk of it for sure. But there is some stuff like that Sambanova or whatever which is a big thing to catch before market. Or the Nvidia news. Whether that translates to a play is another for sure. I think based on criteria or possibly a reaction of the signal in TF could then archive that and basically have Mimir look for any news pertaining to the “rumor” being reported via any of the feeds we have and taking those two points to generate a signal since the rumor is now being reported on?

1 Like
8

Yeah that’s kinda what I’m thinking with the summary function. Basically have it watch everything and notify when there is a drastic change from the current “picture” it has of a company.

9

Just using it to create and basically start possibly tracking a signal. Then when the signal gets reported on boom the rumor is now out to the markets and then the signal is verified and it’s a play is my idea

10

Yes that is a part of the plan, however, being able to detect when rumors are gaining steam is also an important piece as the ideal scenario is identifying opportunities that can be front run. Otherwise we’re doing it all for nothing and should just focus on something that scans the news.

11

Could also go as far as to take that verified signal action/kickoff and then use it to parse the rest of the rumors that are circulating and use that to have a bit of an edge of understanding whether that reported rumor will develop in a particular sentiment direction. That’s definitely an edge in my eyes

12

Yeah that but may require utilizing the rumors from Blind to scrape socials then?

1 Like
13

Yeah I have substantial interest in building out it’s ability to judge how important news is over time and in which direction

Blind is just the test case, in practice the system will be watching Blind, Reddit, StockTwits, our Twitter feeds & our news feeds.

Data from the scans of SEC filings will also be married to the system but will obviously go through a different analysis.

14

Yeah cause if we’re getting conflicting rumors from Blind compared to what’s being reported that could possibly influence some sentiment gauging of sorts. The thing is that those employees on Blind aren’t usually smart enough or high enough to have good material info

15

They may not, but the comments are useful for insight none-the-less. For instance this example from Walmart that it spit out earlier:

Walmart: It appears that Walmart has stopped hiring and has cancelled most contract roles. There is speculation that layoffs may be next, and that most high paid employees are in California. This could lead to a news announcement that could have a material effect on the price of Walmart’s stock.

Individual complaints may not yield much, but when combined across many comments, a picture can emerge. Depending on the prompt, GPT is capable of combining that picture into usable information.

So yeah, while not always going to be a crazy edge, it’s information that could be extremely useful in decision making. But also as we’ve seen it definitely pulls important stuff when it is there.

1 Like
16

Hmmmm yeah it would be interesting to see if we could compile what the companies use as vendors so we would also figure out if a company is going to lose or gain business. I know for compliance we have to list all the vendors in scope for some reports… They may be an interesting way to go about it. We won’t get SOC reports cause we have to sign an NDA but some way of getting what vendors are utilized by a company would be helpful for this idea