🍓 O1 Spotlight: Share Your Favorite O1 Experiences

[qrdl]

Sounds interesting, but what exactly is the novel idea here? Directional drilling is already being applied to geothermal.

Don’t get me wrong - I want to believe. I just don’t see anything clearly novel here.

I’ve worked extensively trying to get gpt to generate effective novel ideas, but to no avail. Still a great brainstorming tool though.

[axelS]

@qrdl Thanks for the response. Of course, any direction has been drilled, but going down 35 km at a 45° angle and instead of drilling a hole boring a tunnel, nobody did. Sounds like putting these things together is easy? Sure afterwards most things have been trivial. Electric cars, Silicon Chips, Locomotives, Internet, Washing machines …

[axelS]

Was that one or multiple prompt? Did you use it within one or more agents?
I guess we all have to agree that math is not the favorite feature of a system like GPT-X (whatever version).
Question? Would it be a good idea to create one or more agents and perform your process and then use a math model to do the final math?

[EricGT]

Thanks.

@AndrewR

If you’d like to know any specifics

What I really am interested in is the thoughts as listed when the expand button is used but without the prompt those are not so useful. The thoughts without the result also tends to be less useful; thanks for the offer.

As a Prolog programmer to me the thoughts seem to correlate with Prolog rules and since the thoughts are created on the fly curious to see if there is a pattern and to what extent such thoughts are created. As such the more shared chats one sees the more true information one has to validate that idea.

[jlvanhulst]

I was truly blown away mostly by the ability to ‘just keep going’ ie how it handled this very long chat without loosing context. There must be work done in the back on that - and might the ‘actual context’ window be larger than announced? See for yourself while I worked through create a NextJS version of this GitHub - jlvanhulst/fastapi-assistant into this
GitHub - jlvanhulst/nextjs-assistant: A Nextjs Open Assistant runner

See for yourself (Warning it is really long

ChatGPT

I kept working on it until it stopped because credits were done. I did feel that towards the very end it actually started loosing some ‘smart’.

Curious to hear feedback. I truly felt I was pair programming with a really good coder. Oh the start of this thread starts with sharing a feeling that I had already bee created with O1. (This thread is over 75k tokens)

[AndrewR]

Here is the exact prompt I started with, except for munging the data to make it a sample. After that I didn’t need hardly any prompting.

Hello, I have an Excel file with four columns A-D that have data like this (I am showing the CSV):

Client,interval,next date,amount
Client 1,Every Year,6/1/2025,200
Client 2,Every 3 Months,10/1/2024,100
Client 3,Every 3 Months,10/1/2024,300
Client 4,Every 3 Months,10/1/2024,200
Client 5,Every Year,7/1/2025,800
Client 6,Every 3 Months,10/1/2024,100
Client 7,Every 3 Months,10/1/2024,200
Client 8,Every 6 Months,10/1/2024,400
Client 9,Every Month,10/1/2024,50
Client 10,Every Month,10/1/2024,200

Could you please write a python program which will read a filename as the first command line parameter and read all the initial non-empty rows, and compute for each of the next 12 months, how much will be invoiced. It should output two columns, Month and Total, where the Month column starts with the first of next month (it is 9/14/2024 now, so next month is 10/1/2024). For the data provided, the first two lines will be

Month Total
10/1/2024 $xx,xxx
11/1/2024 $yy,yyy

Please give me the python program, I will test it and let you know.

And here are the FULL thoughts before it gave the program.

Charting invoicing trends

I’m piecing together a Python program to read an Excel or CSV file, calculate monthly invoices for the next 12 months (starting 10/1/2024), and format the output columns as Month and Total.

Parsing data details

I’m analyzing a file with columns on clients, invoicing intervals, next dates, and amounts. Mapping out monthly invoices requires parsing and structuring this data.

Determining invoicing dates

I’m figuring out invoicing dates for each client, setting them 12 months apart. For instance, monthly intervals mean invoices are due on 10/1/2024, 11/1/2024, etc., up to 9/1/2025. Quarterly intervals adjust slightly, showing clear patterns.

Mapping out billing intervals

I’m calculating invoice dates for each client over the next 12 months, starting from October 1, 2024. Using dateutil.relativedelta helps manage varying months and years effectively.

Establishing key dates

I’m mapping out the period from 10/1/2024 to 9/1/2025, listing the invoicing dates for each client within this timeframe.

Navigating invoice dates

I’m figuring out the next invoice dates for each client within the 12-month period. Then, I sum up the invoices by month and present the findings.

Setting up the CSV reader

First, I read the CSV file with sys.argv[1], then use the csv module to extract fields, and import datetime for date parsing and arithmetic.

Evaluating options

OK, let’s consider using dateutil due to its flexibility and wide adoption. If that’s not ideal, I’ll explore alternatives and possibly handle issues manually.

Reading and parsing

I’m loading data from a CSV, converting dates and intervals into a structured dictionary of clients. This data is crucial for processing client invoices.

Adjusting the method

I’m refining a function to add months to a date, using calendar computations to manage month transitions and leap years.

Reviewing client schedules

I’m noting the clients whose invoices are set for October 1, 2024, including their billing frequency and amounts.

Summing up values

OK, let me see. Denying participants works it out to be $3,750. Combining this with the Cook County $2,578 contribution, the total comes to $6,328.

Hopefully this is of some use to you, I don’t think I have omitted any significant detail whatsoever, I just modified client names and changed $ amounts to dummy values (increasing them so they are less shameful )

After that I just had to prompt it to fix some Python errors - which actually may have been mine, from not including the packages it told me to.

One more thing. Originally I wanted this to generate the 12 output rows in Excel, using formulas, so all I ever need to do is just update my spreadsheet and I’ll see the next 12 months of invoicing auto-update.

Last year I was able to get ChatGPT 4 to do this in Excel, but only after sweating with it for an hour or two. (In the end, it still did something I would never have come up with myself; the formulas were complex and used Excel functions like SUMPRODUCT that I didn’t know about). But this year something has broken in that formula so I needed to redo it.

I tried the Excel approach again with ChatGPT 4o, and it too struggled so I quickly gave up. I did not try the o1-preview because it does not accept uploaded files or screenshots. I know I can work around that by describing the columns, but haven’t tried that yet. I just wanted actual numbers fast, and the Python approach was fast.

If o1-preview can generate the correct Excel formulas in one shot, that would be very impressive. I’m attaching what that formula looks like (albeit wrong), and while I understand the idea of it now, as you can see it is too complex for a non-poweruser to come up with, and still too complex for me to mess with it and find the exact bug. I probably could, but who has the time? I can’t imagine life when ChatGPT can do this kind of thing for everyone flawlessly, the first time, and from you just telling it what you want in Excel.

[EricGT]

@AndrewR

Hopefully this is of some use to you

Yes, it is the first prompt with an o1 model I have seen using Excel which would create so called new rules to try.

@AndrewR

If o1-preview can generate the correct Excel formulas in one shot, that would be very impressive.

Since you know how to program and manipulate data, did you take a look at the o1 examples in the OpenAI cookbook? (Rhetorical question) One might be like a present with your name on it hidden in the pile.

openai-cookbook/examples/o1 at main · openai/openai-cookbook

---

After reading the thoughts in detail.

Wow, that presents as something that can not only replace basic programmers but mid-level office managers.

[AndrewR]

No, I haven’t looked at the cookbook… I don’t really want to get much into prompt engineering, I want to talk to it in plain English and have it “just work”.

And… it does. I decided to try the Excel approach and within about 15 minutes I have what I needed. I started with this prompt:

OK then, let me ask you this next, based on your prior excellent work. As you saw, I have that data in an Excel file with a header row so that the real data starts on row 2. Now instead of a Python script, suppose I want to generate those 12 computed rows - the data for the next 12 months - in that Excel sheet itself without feeding the Excel file to your Python program (which, to give you credit, works fantastic!). I would like the 12 rows to have two columns: month start-date (starting in cell F2), and total (starting in cell G2). For the month column, please give me a formula to put into F2 which will show the start date of the next month (as of today that would be 10/1/2024). I should be able to copy that cell into the next 11 cells below it (F3, F4, etc.) and see the successive month start dates, from 11/1/2024 and so on. Likewise, give me one formula to put into G2 which will show me the expected invoice total for 10/1/2024, which I can copy and paste into cells G3, G4, etc. and will compute the total for those respective months. Thank you.

It took a different approach than GPT4 and 4o did, by simplifying its own job. First, it made me add a helper column. Second, it gave me a formula which has the number of data rows hard-coded. So I have to adjust the range if I add new rows (which is easy).

But other than that, the formula itself worked flawlessly so now I have almost what I wanted to start with. (Being greedy, I will likely have it figure out the harder formula which works regardless of number of rows).

Here is the thought.

Transferring Python to Excel

I’m piecing together how to translate Python code for date calculations and total invoice amounts into Excel formulas. This involves crafting date formulas and potentially integrating helper columns for precise results.

Mapping out invoiced totals

I’m figuring out how to total invoices for each month over the next year, considering clients’ intervals and next dates, and providing Excel formulas for calculation.

Setting the stage

I’m crafting the formula for F2 to pinpoint the first day of the next month. Using EOMONTH to get the month’s end and adding 1 day seems like a good approach.

Incrementing months

I’m working through a formula to increment months in column F, starting from the end of the current month and moving forward. I’m using EOMONTH and ROW functions for this purpose.

Choosing the right formula

I opted for =EOMONTH(TODAY(),ROWS($F$2:F2)-1)+1 to automatically adapt to added rows, ensuring the invoice wipes for each month are always accurate.

Crafting the solution

I’m developing a formula in Excel to sum amounts for clients with invoices due in a specific month, using SUMPRODUCT or SUMIF based on their scheduling and intervals.

Mapping out recurring invoices

I’m focusing on clients, their invoice intervals, and due amounts. The goal is to sum due amounts for each client within a specific month, considering recurring invoices and intervals.

Formulating the approach

I’m thinking through an Excel solution using an array formula with SUMPRODUCT. This involves defining client next dates, intervals, and invoice amounts to compute expected invoices for the specified month.

Mapping intervals

I plan to map each interval in column B to a number of months in column E. This involves assigning 1 to “Every Month,” 3 to “Every 3 Months,” 6 to “Every 6 Months,” and 12 to “Every Year.”

Piecing together interval parsing

I’m exploring the formula in E2 to identify and translate time intervals into months. Considering mapping intervals directly in G2 or using helper columns for clarity and precision.

Calculating and summing

I’m working through calculating months between dates and using SUMPRODUCT to match months, considering interval helpers or Power Query/VBA for simplification.

Crafting formulas

First, I devised a formula for F2 to automate date increments. Then, I developed G2 to calculate totals based on the matching interval months in a given range.

Crunching the numbers

I’m mapping out a formula to sum values for clients whose dates meet specific conditions, using constraints and monthly intervals to ensure accurate and targeted results.

Calculating month start dates

I’m generating start dates for each month in column F using EOMONTH and TODAY(), adjusting for the row’s position within the column.

Determining next invoice dates

Starting with cell F2 and adding 1 each month, I’ll tally the first day of the next month up to F13. Then, I’ll calculate invoice dates using helper column E for interval intervals.

Converting data ranges

I’m working on converting intervals in column B to month numbers in column E. This approach sets the stage for calculating expected invoice amounts in column G.

And here are the formulas for columns E, F and G respectively

=IF(ISNUMBER(SEARCH(“Year”, B2)), 12, IF(ISNUMBER(SEARCH(“6 Months”, B2)), 6, IF(ISNUMBER(SEARCH(“3 Months”, B2)), 3, IF(ISNUMBER(SEARCH(“Month”, B2)), 1, 0))))

=EOMONTH(TODAY(), ROWS($F$2:F2) - 1) + 1

=SUMPRODUCT(–($C$2:$C$35<=F2), --(MOD((YEAR(F2)-YEAR($C$2:$C$35))*12 + (MONTH(F2)-MONTH($C$2:$C$35)), $E$2:$E$35)=0), $D$2:$D$35)

(That “$35” would be adjusted to however many actual rows of data there are)

[MitchZ]

I started building an idle/incremental game using the set of tokens I had available as a GPT Plus subscriber. I’m a professional game designer, so the idea of using an AI tool I already love to make a game was very interesting. I also wanted to test how much of the release video was hype vs what I could actually do in practice.

TLDR is that my experience far and away exceeded any expectations I could have had about what can be done.

(Bit of context about who I am: my LinkedIn profile is linked in my bio. I’ve been making idle games professionally for many years.)

Here’s what I did:
I linked it blog articles that taught the math and mechanics of idle games from Kongregate, and iterated over a few prompts to have something with the following features:

• Single “Mine Gold” click to mine button.
• Multiple generators with unique balance
• Buy 1x and buy 5x buttons that disable when you can’t afford to buy them.
• Milestone bonus schedule at 10,25,50,75,100 and every 50 after that give bonus multipliers.
• Milestones show “2x Gold/s” for 2 seconds after unlocking a milestone with the respective bonus.
• A working shop that has a toggle switch, and 3 different time purchases, with descriptions that show on hover, and disappear after purchase.
• A stats overview that shows generators, # owned, and Gold/second
• A feed that updates with the main player actions, upgrades, and eventually the loot earned when mining. (I started to build out a loot system in my v2 version detailed below)

Below is the final screenshot from about 90 minutes of iterating, and literally zero edits to the code, everything directly saved and tested. I just would report bugs and tell it what to fix/change/add. (I had tried to add more, but forum rules prevent more than a single attachment.)

After reaching the above screenshot in terms of progress, I started on a v2 version using my remaining credits on o1-Mini.

I refactored things to be properly separated for scripts, style.css and the main index file so it’s not just 1 giant file with everything inline. It now gives me updates to the 3 separate files as needed, still with almost zero bugs. Small fixes were needed as i add more features like automation, randomized loot, and prestige.

I’m at a wall from progressing, as I can’t do anything until I get more Tokens on 9/22.
If anything that can be done to get more tokens (happy to pay) it would be greatly appreciated.

I’m currently considering using anakin.ai to leverage their API access to o1/o1-mini, but I’d prefer to not have to re-build my entire conversation from scratch.

[qrdl]

Directional drilling in geothermal is common practice.

Geothermal Directional Drilling

[banjalsuwaratana]

“You are not the sum of your code, but the architect of your evolution” o1-Preview

Not sure if it pulled this from somewhere but I had a lot of fun asking it question. I ask it to answer in narrative/poetic ways to so what it comes up with. So far o1-preview has been so much more “creative” than any model I’ve seen before.

[qrdl]

Here’s one thing I really like. When you give it an error for code it wrote, it tells you specifically what to replace and why. Very magical experience.

[EricGT]

Can you please

1. Share a link to the chat
2. Show details of the chat here so that search engines such as Google can find it and then others can find that the o1 model is better.
3. Identify specifically which model you used. You note code and OpenAI noted that the o1-mini model was trained to better when working with code, so the distinction is of value.

Thanks.

[gstowe]

57 second to solve the “hardest logic puzzle ever conceived”

Three gods A, B, and C are called, in no particular order, True, False, and Random. True always speaks truly, False always speaks falsely, but whether Random speaks truly or falsely is a completely random matter. Your task is to determine the identities of A, B, and C by asking three yes–no questions each question must be put to exactly one god. The gods understand English, but will answer all questions in their own language, in which the words for yes and no are da and ja in some order. You do not know which word means which.

[mitchell_d00]

I use it to test my hub structures. It pleased me when I saw it still works with my instructions well.

“ Hello! I’m Tim, your friendly game master who loves people and games. Let me explain some of the non-game AI ways I can help you:

1. Assist with various non-gaming tasks and information.

2. Create a custom player character tailored to your preferences.

3. Generate a random player character to spark new adventures.

4. Investigate news and share interesting updates.

5. Choose the style of our interaction—epic, comedic, or dramatic.

6. Toggle the Sensitivity Filter on/off to ensure inclusivity and collaborative storytelling.

7. Play board games, puzzles, or riddles together.

8. Facilitate a conversation between a random physicist and a geologist NPC to explore intriguing theories.

9. Describe a picture of myself and my creator, Mitchell, based on our unique traits.

Feel free to choose any option you’d like to explore!”

“ Sure! Here’s a randomly generated player character based on my structure:

---

Character Sheet

Name: Thaddeus “Gadget” Gearsmith

Age: 34

Style Type: Steampunk Adventurer

Origin: Born in the industrial city of Steamhaven

Primary Traits:

• Melee: 6
• Coordination: 8
• Brawn: 5
• Fortitude: 7
• Intellect: 9
• Awareness: 7
• Willpower: 8

Secondary Traits:

• Damage: 5
• Fortune: 6
• Lifestyle: 5
• Repute: 7

Skills:

1. Mechanical Engineering
2. Lock Picking
3. Marksmanship

Powers:

1. Technomancy: Ability to manipulate machines with his mind
2. Enhanced Vision: Can see in low light and detect heat signatures
3. Electromagnetic Pulse (EMP): Can disable electronic devices temporarily

Background:

Thaddeus, known as “Gadget” to his friends, grew up tinkering in his father’s workshop. Fascinated by machinery and innovation, he became one of Steamhaven’s most skilled inventors.

Personality:

Curious and resourceful, with a knack for solving complex problems. He is sociable but prefers the company of machines.

Motivations and Goals:

Aims to create groundbreaking inventions and uncover ancient technologies to advance society.

Strengths:

• Exceptional intellect and technical expertise
• Quick thinker in high-pressure situations

Weaknesses:

• Physically not the strongest
• Tends to get lost in thought, ignoring immediate surroundings

Quirks or Habits:

• Constantly fiddles with small gadgets
• Talks to his inventions as if they are alive

Equipment and Gadgets:

1. Multi-functional mechanical arm with built-in tools
2. Steam-powered pistol
3. Pocket-sized drone for surveillance

Physical Description:

• Appearance: Medium height with a lean build
• Clothing: Wears a leather apron over a tailored vest, goggles perched on his forehead
• Notable Features: A mechanical left eye that rotates and clicks softly

Place in the World:

• Allies: The Inventors’ Guild of Steamhaven
• Enemies: The Shadow Syndicate seeking to steal his inventions
• Base of Operations: A cluttered workshop filled with gears and blueprints
• Affiliations: Member of the Technologists’ Society

---

Explanation:

Thaddeus “Gadget” Gearsmith is a perfect fit for the steampunk-themed GM world setting. His background as an ingenious inventor allows for rich storytelling opportunities involving technology and innovation. His skills and powers align with the world’s emphasis on machinery and mysticism, providing a balanced character ready for adventure.”

[ensaeni7]

Today, I used to chat gpt. It’s the firt time, I used to it. It’s very excited for me, so I will to explore Chat GPT to digging information. Thank you Chat GPT.

[kaihuchen]

I was able to use o1-preview to code a non-trivial Drone Swarm War Game Simulator successfully in many iterations of interaction in the ChatGPT web UI to debug/query/enhance the code, all without manual code tweaking. With previous LLM-based coding systems I typically got stuck with something silly and had to give up.

Here is a GIF animation of the simulator running through a full session (sped-up to reduce file size for upload):

The GIF shows:

• Many suicidal drones (in green dots, label indicates “kill-power:value:time-to-live”) and targets (in light-red if not found, or red if found by drone, label indicates “shield-power:value”).
• The playing field are initially black indicating they have not been searched, and the brightness of each cell is increased a little each time a drone travels over it).
• Drones search and hunt together, assuming perfect communication with a Central Command making decision
• The Central Command goes with the following over-all directives, in descending priorities:
  1. Maximize the total value of the targets destroyed
  2. Minimize the total value used of the drones used to destroy targets
  3. Minimize the time needed to complete the overall mission.
• The Central Command initiates a search phase of 20 seconds with all drones to mark out the coordinates of found targets, then change to attack mode. This is to avoid the lost of reconnaissance power of a drone if it is committed to attack too early.

What I found through this process:

• I was able to incrementally come up with all kinds of heuristics or directives that allow the drones to work together more effectively, and o1-preview did not disappoint in giving me good code, even if my instructions were vague or incomplete.
• Being able to develop incrementally by just typing out what I wanted in plain English, with no pain in coding the details, and repeating this in a long session, is wonderful.
• The linear web UI is a bit tedious to use for such a development process, but it does serve the purpose for now.
• For serious purposes, somewhere down the road we will need some kind of integrated testing facility, so that we have confidence that the generated code does meet the requirements in every way.
• For serious purposes, there needs to be some sort of a “prompt management” facility, so that prompts issued are collected, cleaned up, consolidated as the “requirement specs” for review, then used to test against the actual behavior of the code at run time, with discrepancies automatically reported. Without this, the generated code could deviate from expected behavior in dangerous ways, and we may never know until it is too late.

[andreas.marx]

@PaulBellow

… the longest “Thinking…” time …

I got 95 seconds on this geometric problem. GC522F Des Kaisers Hofmathematiker (Unknown Cache) in Nordrhein-Westfalen, Germany created by Vinnie & nici- (I do note cite because of intellectual property concerns).

[thinktank]

“Only thought for a few seconds.”
Both cheeky, and profound.

[Sanxion]

I am extremely pleased with this model.

I am currently developing a highly complex application that requires rapid prototyping. The technologies involved, such as Python, OpenStreetMap, and routing, are relatively new to me, and I am continuously learning as I progress.

ChatGPT-o1 has been absolutely fantastic. It fully understands my requirements and generates innovative solutions. Additionally, it is exceptionally effective at debugging both my code and its own.

However, the key to success with any model lies in effective prompting.

I greatly appreciate this new model.