I know the list may be long, but honestly ChatGPT Pulse regularly finds items I am interested in that I may not find otherwise. Also I have been tweaking and curating ChatGPT Pulse for a few weeks so it may seem like it just works; it does take some effort, like prompt engineering, but worth the effort.
Releases · leanprover/lean4-nightly · GitHub
Notice the URL ends with utm_source=chatgpt.com
It is nice to get updates on public code changes. Saves me lots of time manually working through a list, and ChatGPT Pulse gives a brief summary so I don’t have to read as much and know if diving deeper is worth the effort.
UI hooks: In your Cytoscape view, surface these four properties as live checkboxes/badges per graph; when all four light up, enable “Export Cayley table,” then run your existing property tests. (Nice fit for your messageBus + property panels.)
Currently doing my usual morning current events reading.
This paper popped up on one of the list, curious to see if ChatGPT Pulse adds it today or maybe later this week to one of the weekly repeating sections. If not will have to curate ChatGPT Pulse to find such.
“Quantifier Elimination Meets Treewidth” (PDF) by Hao Wu, Jiyu Zhu, Amir Kafshdar Goharshady, Jie An, Bican Xia, and Naijun Zhan
Here is an image that ChatGPT Pulse made for Practical ways to cut Windows build time
Sorry to report that links for arxiv and GitHub in the section Fresh research roundup are resulting in 404; the title of many are not even close enough to find the items. So an entire section of 18 items of no value. Note: The invalid links are only happening in one section, the links in the other sections are working correctly, even for arxiv and GitHub.
While there were other items of value, they were not related to programming.
Since I can not post more than 2 replies in a row will have to post this reply here.
Reading my ChatGPT Pulse update for January 6, 2026
Feature Proposal: Agent Continuity - Persistent Agents Across Session Boundaries
What is nice about ChatGPT Pulse is that there are 5K+ Issues for this GitHub repo, this one was opened 18 hours ago, and it was selected based on my interest such as avoiding chat session compaction and resuming without using tokens to figure out the current state.
The broader context here is that the AI ecosystem — from tools like Model Context Protocol (MCP) for context integration to community‑built plugins tackling session continuity — is converging on memory and continuity as first‑class design goals for multi‑agent systems.
Reading my ChatGPT Pulse update for January 7, 2026
One of my weekly updates included
“Application of Large Language Models for Annotating Genes into
Reactome Pathways” (PDF) by Guanming Wu, Lisa Matthews, Nathan Boyer, Marija Milacic, Deidre Beavers, Nancy T Li, Bruce May, Karen Rothfels, Veronica Shamovsky, Ralf Stephan, Marc Gillespie, Henning Hermjakob, Peter D’Eustachio, Lincoln Stein
Reading my ChatGPT Pulse update for January 8, 2026
Maybe the OpenAI Devs are reading this topic because today the links are correct, e.g.
When ChatGPT Pulse is open, you will see in the upper right a Curate button,
using that sent this prompt.
The Consumer Electronics Show is happening. There will be lots of announcements. Send me updates every hour and I will use the up/down thumbs to let you know how to adjust as needed.
Saw this earlier today with Google News, still nice to know ChatGPT Pulse also surfaced it for me.
Noted this to ChatGPT Pulse
For the next CES hourly update, can the bullet point items include a thumbs up/down so I can refine at that level. The first one for Nvidia gets a thumbs up, the rest are fine, and no thumbs down.
which responded with
Acknowledged.
Next CES hourly update format
Each bullet will include a / selector for item-level feedback.
Your current preferences are recorded as:
Nvidia announcements (high priority)
Others: acceptable baseline
No exclusions or down-ranking applied
What this affects
Ordering: Nvidia-related items will surface earlier.
Coverage depth: more technical detail for Nvidia (architecture, benchmarks, roadmaps).
Scope: unchanged for now unless you add feedback later.
The next update will follow this format so you can adjust relevance at the bullet level.
Tried one more time to get ChatGPT Pulse to use some sort of clickable feedback for each item and failed.
Looked for a way to report this on OpenAI GitHub or OpenAI ChatGPT Pulse FAQ and found none. Will have to look further later.
Lego unveils Smart Play Bricks
This item linked to a paywall.
Would be nice if there were a way to ban paywall sites and have the AI search for the same information directly from the source, e.g. LEGO® SMART Play™ System
Yesterday ChatGPT Pulse stopped updating every hour after the 5th version. Really did not mind because it was surfacing nothing new and some of the earlier replies included sites that were doing regular updates
Posting this here to avoid the more than two replies in a row limiation.
Now that CES is over, changing gears to something new
Using the curate feature
Quiz me daily on
* SWI-Prolog
* Prolog programming
* Lean 4 programming
* Lean 4 proofs
* Python programming
Also ask for more specific feedback than just using the thumbs up/down with card interactions. While effective thumbs up/down are not fine grained enough.
That does not resemble any ChatGPT Pulse update I have seen. I do not see any external links.
ChatGPT Pulse updates typically begin with a notification banner at the top of the list, for example:
In your link this might be considered the banner at the top
but the long list looks more like a plan that has gone to an extreme.
All ChatGPT Pulse results I have seen are presented with the URL https://chatgpt.com/pulse, if you do not see that then it may not be ChatGPT Pulse. I have not seen the results presented via any other means, the notifications do come in via multiple means but those link back to the URL.
It might still be in beta release. From my use and what is noted by me, there are still rough edges, but for me the glass is half full.
Since I am not an OpenAI employee can only present what I know which is what the FAQ notes about only being available for Pro users and limited to certain means.
Reading my ChatGPT Pulse update for January 12, 2026
This problem is back.
Many developers using AI tools seem to experience the same challenges at some point; I have encountered this myself with several different tools.
I would not normally spend time looking for this kind of information, but knowing it is a common issue makes the experience easier to accept.
This was a link to a site where I had to watch an ad only to see something I check regularly and do not have to wait for an advertisement and get it straight from the source.
the source of truth link is here and free
Might have to curate ChatGPT Pulse to return source of truth when possible.
The official MCP Registry has become a go‑to source for discovering actively updated MCP servers like the Appium MCP entry, which was refreshed on January 11, 2026 — great for finding vetted, interoperable tool endpoints that agents and IDE clients can plug into. You can explore this catalog directly and pull server metadata programmatically.
So used the link from ChatGPT Pulse to check on this.
Discourse the software this forum runs on, registered their Discourse MCP server with the registry.
Now that CES is over wanted to delete the CES task.
In took a minute to figure out that the “X” only closes the window, you need to use the trashcan.
Click “Curate”
Select the desired task and click it to open
In the upper right click on the trashcan
Example of my first daily quiz question grading.
1. SWI-Prolog: current_predicate/1 vs current_predicate/2
Your answer (summary):
Correctly identified autoloader interaction as a performance issue.
Noted failed autoloader lookups being expensive.
Uncertain about the tooling pitfall.
Assessment:Mostly correct, but incomplete
What’s correct
current_predicate/2can consult the autoloader.
Autoloader index verification is a real cost and can dominate runtime.
What’s missing (the key pitfall)
Tooling code that merely inspects predicates may accidentally trigger autoloading, changing the program state.
This can:
Load modules you didn’t intend to load
Introduce predicates that were not previously present
Make results non-reproducible depending on call order
Canonical pitfall phrasing
Using current_predicate/2 in analysis or tooling code can have observable side effects (autoloading), violating the expectation that introspection is pure.
Score: 7.5 / 10
Here is a nice ChatGPT Pulse feature hidden in plain sight.
Right under the title is a bookmark/save icon to save as chat.
What is really interesting is that the paper notes
We also present an automated pipeline that leveraged OEIS-LT to formalize more than 25,000 OEIS sequence definitions and prove more than 1.6 million theorems.
Thank you for the reply. Definitely extreme and unnecessary. Updates and resend every week. Appreciate the incite and the thread. Will keep investigating.
@RiavvioAS
I just double-checked availability, and it appears to be limited to Pro subscribers for now. Sorry about that. Hopefully more compute capacity comes online so this feature can be expanded more broadly.
@EricGT
This feature has been enabled on my account for a while, but I painted myself into a corner. The daily suggestions are still focused on items I asked to be curated one or two months ago, and I never updated them.
After you started this conversation, I kept thinking about how this could deliver real value. As I increasingly use AI to do work for me, there is a risk of unlearning certain skills, or at least losing touch with the underlying work and reasoning needed to produce good results. I want to try using Pulse as a way to reinforce, revisit, and better understand what I largely automated the previous day with ChatGPT.
For example, if I asked about taxes in my country yesterday, it would be useful to receive a short quiz the next day to confirm that I actually understand and remember what we discussed. That kind of approach might help me use the tool more consistently and get more value out of it.
Ps. just updated the task. First actual learning of today: We can define ourselves at what time of day we receive the daily Pulse. That’s good to know! I may just try putting it into the evening as a “reflect on the day/plan the next day” activity.