Jarad Johnson

@deanwball I’ve been very surprised by the lack of progress with Pulse. A daily briefing is very powerful and could be extended through apps/connectors to do real work overnight. Also surprised Claude hasn’t entered the space.

I hope that in the “refocusing” OpenAI does not drop Pulse, which I find insanely useful for surfacing important but under-the-radar news items almost daily.

Interested if OpenAI sees Codex as its Claude Cowork response. I only know one non-technical person using Codex. The rest are in love with either Cowork or Perplexity Computer.

@ClementDelangue I think this is one of the worst use cases for generative AI. Can be helpful for taking brain dumps and organizing them, but AI generated pre-written replies just remove humanity and cause confusion.

why don't all messaging apps (slack, whatsapp, gmail, linkedin,...) have great AI powered suggested replies and great autocompletes? Sounds kind of trivial given they have your voice and history and would save everyone tons of time. Too costly maybe with big models (😉😉😉 smaller open-source on-device models)?

@JacobColling FWIW Granola's founder seems to get it (full thread): x.com/cjpedregal/sta… I'm an advocate for Granola having used quite a few of these. It's the best I've used and the MCP does what I need to work with agents. API would be great for special use cases, though.

I was just going to re-sub to Granola but reconsidering after seeing this. Best note taking app that plays well with agents?

@_simonsmith Do you use GPT-5.4 frequently in Codex as well? I find it to be less verbose there and capable of matching Opus. But still prefer Opus in the main apps.

I go back and forth between GPT-5.4 and Claude Opus 4.6. The former feels a bit smarter and more accurate with search, but can be too verbose. The latter is way better at design and a bit more personable, but can be sycophantic ("great question") and naive with search (e.g. giving answers that are obviously outdated based on what it finds, without critically assessing the search results). If the next GPT has massively improved design skills and tighter writing, and the next Opus dials back the sycophancy and elevates reasoning and critical thinking, the model families will converge and usage will likely come down to switching inertia and non-model preferences like harnesses, UIs, and personality.

We're road tripping for spring break and the random Texas town we stopped in last night had an annual event like this going on. This version of America is everywhere unless you actively avoid participating in it.

Every time I see these posts they are just from people who seemingly make no effort at all to experience community.

"The sooner David Ellison takes over that network, the better." The Defense Secretary calling for a billionaire ally of the government to take over a media outlet & make it more pro-government. This is what authoritarianism looks like. But in other countries they try & hide it.

Both models do this. GPT-5.4 skipped a step in my day recap process yesterday. When I asked why, it told me I needed rest.

@JaredSleeper Both models do this. GPT-5.4 skipped a step in my day recap process yesterday. When I asked why, it told me I needed rest.

One of Claude’s best features is that it will often tell you to “go to sleep.” Can’t imagine chatGPT doing that- it acts like it is desperate for every last second of engagement. It’s so clear which one is healthier/better.

@ycocerious Pushups. jaradjohnson.com/writing/the-pu…

What do you guys do while your claude code is running?

Codex is back to hanging consistently.

@_simonsmith I've read some with this experience. It has improved for me over 5.2, but inconsistently. What led to my post was asking it why it wrote slop and it replied that it prioritized writing a "strategy document" over using my voice DNA. I may need to reshape my skills.

@jdjohnson I don't find this. If I tell it "polish my writing but do not change my meaning or voice or style, and don't turn this into generic AI output or slop," it does a really good job for me. Better than Opus, I find, as it tends to over-edit.

GPT-5.4 still struggles with prioritizing how it thinks something would be written over how the person actually writes. Even using a writing skill, if it's editing for a specific use case, it prioritizes that use case (e.g., a social post becomes on sentence per line slop).

@_simonsmith @signulll Best use case by far in ChatGPT. I do it weekly for a new focus playlist in different genres/time periods.

@signulll Only for music. "Make me a playlist that's <very specific direction>." That works well.

i’m curious, do ppl use apps inside of chatgpt? if so, which one & what’s the use case? i haven’t explicitly used an app at all yet nor have i come across gpt prompting me to use one. also the directory doesn’t seem updated since launch.

@signulll Up until GPT-5.4 they failed on about 80% of uses, so it wasn't practical. But Claude shows there's decent use cases. For example, a salesperson could have it check their calendar, then provide a brief for any external meetings with additional context from email/CRM/search.

@v_lugovsky I still had to do a lot of follow up work, but it laid the ground work and what would have taken several hours ended up taking the first hour of the day.

@v_lugovsky For sure. Last night mine ran 25 steps. Recaps of sessions for our consulting clients, an RFP analysis, a science lab for my kids we'll run tonight, health research for my 4-year-old, and a full morning briefing for all my tasks organized by priority.

I built my own ChatGPT Pulse with my local agent. It's much better. And proves to me that OpenAI was onto something with the feature. It does real work every night I can use the next day. They really should give the feature more attention.

@_simonsmith Any idea why GPT-5.4 also has a consistently bad web design aesthetic? In that case, it does seem like it has been given instructions, because the results are similarly bad across different projects. OpenAI clearly cares about design, but it doesn't come through in the models.

So, ChatGPT team responsible for presentations: the Slides skill (at least in Codex; assuming also ChatGPT) is undermining GPT-5.4's ability to create good presentations. Opus helped me diagnose some issues: 1. It's using LAYOUT_WIDE as the default, which is 13.33" × 7.5", instead of a standard 16:9 of 10" × 5.625". 2. The helpers and authoring rules create too much complexity. If I want a few slides, it has to go through so much complex work, which takes a long time and doesn't seem to help. Opus notes: "There are 2,371 lines of JavaScript helpers... The skill tells ChatGPT to copy this entire folder into the workspace and use it. For a 20-slide data-dense deck this makes sense. For a simple slide, it's like handing someone a CNC machine to cut a piece of paper." Good one :) 3. There's no design guidance. According to Opus, "The word 'design' doesn't appear once in the SKILL.md. No color palette suggestions. No typography pairing advice. No layout principles. No guidance on restraint or whitespace. No examples of what 'good' looks like versus 'bad.'" Some recommendations: - Let ChatGPT work freely in PptxGenJS when warranted, such as for simple slides. - Add some design principles and guidance - Change the default layout to LAYOUT_16x9 - Add 2-3 example outputs as visual references