Sean Murphy

629 posts

Sean Murphy banner
Sean Murphy

Sean Murphy

@AmbRealismX

Turning bold ideas into things that actually work. AI strategy, systems thinking, and execution-first consulting for founders and builders.

Gloucester, MA Katılım Nisan 2025
285 Takip Edilen85 Takipçiler
Sabitlenmiş Tweet
Sean Murphy
Sean Murphy@AmbRealismX·
It's a pretty cool feeling to release my first OSS project forked from one person that influenced me: @theo to include another incredable project from another who influenced me @badlogicgames and then use my "quick" reflexes to capture a repost from the creator of Pi!
Sean Murphy tweet media
English
1
0
3
2.6K
Sean Murphy
Sean Murphy@AmbRealismX·
@NousResearch If only they had a visual way to demonstrate things. Almost like if there was a series of pictures but moving fast enough so that it looks real. Hey Hermes I have an idea!
English
0
0
5
850
Nous Research
Nous Research@NousResearch·
Hermes Agent now comes preinstalled on rabbitOS
rabbit inc.@rabbit_hmi

rabbitOS 2.3 is here, with hermes agent 🥕🪽 a fresh OTA is rolling out to r1 now, and this one’s packed: hermes agent, proactive rabbit, openclaw v4, creations gallery 1.5, DLAM BYOK, and more. here’s what’s new: 🪽 hermes agent on r1 you can now talk to hermes agent by @nousresearch directly from r1 and get things done. simply install the rabbit agent to your computer from rabbithole > nodes and then swipe left on r1 until you reach the hermes page to connect. 🆕 DLAM moves to BYOK (Bring Your Own Key) DLAM now uses your own Anthropic or OpenAI API key. it remains available at no extra charge from rabbit, with usage handled through your chosen provider. check out the DLAM settings page to choose your provider and model and enter your API key. 💥 openclaw protocol v4 r1 now supports openclaw v4, with the integration rebuilt through rabbit agent for a stronger foundation going forward, including usage outside of your home network. 🖼️ creations gallery 1.5 editor’s picks, popular creations, QR codes, intern session shortcuts, and the full creations gallery — all in one cleaner on-device experience. ✨ proactive rabbit triple-tap the rabbit on your home screen and r1 will greet you with something helpful, personal, or just a little fun — based on your r1 personality, reminders, journal entries, magic recorder logs, and more. 🟥 what’s new? a new on-device card lets you see recent OTA updates right from r1. rabbitOS keeps evolving fast. update your r1 and have a play 🧡🥕

English
84
90
1.7K
168.5K
Ksenia Moskalenko
Ksenia Moskalenko@kseniam0s·
Give me a startup name SO COOL it sounds like the next unicorn 🦄
English
693
3
348
109.6K
Sean Murphy
Sean Murphy@AmbRealismX·
When did one shot become the metric. Did they one shot the Sistine Chapel. Stop promoting laziness. Minecraft clones and Rubik’s cubes aren’t the benchmarks that are meaningful. One shotting a prompt isn’t the measurement of quality and it doesn’t make you good at development. Stop promoting laziness. You run a company did you one shot your way into that position? Didn’t think so.
English
0
0
0
14
Merlin
Merlin@Merliiin__·
@DavidOndrej1 You seem like an OpenAI hater for now - are there specific reasons why you dislike GPT so much?
English
1
0
2
150
Sean Murphy
Sean Murphy@AmbRealismX·
@goof @thsottiaux @Youssofal_ When I said look left I mean along a horizontal line. You may need to go up a little to the closest thinking level just don’t go below the invisible line.
English
0
0
0
4
Sean Murphy
Sean Murphy@AmbRealismX·
A good strategy is to look at the chart below. Use Sol for planning on high or Xhigh. Then for implementation choose your thinking level you would use for Sol and look left to Terra and Luna. Find the closest thinking level for those two models and choose. The further left you go the more you save and the faster it goes. If you have a good plan all three models are more than capable of implementing. Then use Sol as a verification step at the end.
Sean Murphy tweet media
English
1
0
0
15
Youssof Al Toukhi
Youssof Al Toukhi@Youssofal_·
My experience with GPT 5.6 Sol summarised in one image:
Youssof Al Toukhi tweet media
English
94
43
1.4K
142.1K
Sean Murphy
Sean Murphy@AmbRealismX·
Too many people are hyper focused on Sol. The real game is figuring out how and when to use Terra and Luna. Look at the chart attached and look at every spot on the Sol line below High. Then draw a horizontal line to the left. Challenge yourself to use 2 models at a minimum, then do the same thing with all 3 being required. Figuring out which, where and when is the whole game.
Sean Murphy tweet media
English
1
0
4
440
Anthony Kroeger
Anthony Kroeger@kr0der·
it's been 1 day now - what reasoning level is everyone using for GPT 5.6 Sol?
English
283
5
291
68.6K
Sean Murphy
Sean Murphy@AmbRealismX·
Don’t sleep on Luna Max and ultra for implementation. If you have a good plan use Sol like the bread of a sandwich. Plan is the bottom piece and pre-PR check is the top slice. The meat can be combos of Terra, Luna & Sol and lower thinking levels. I have been having Sol make recommendations for what models work for particular tasks in the plan and it makes solid recommendations if you tell it the range to choose from.
English
0
0
2
17
David Ondrej
David Ondrej@DavidOndrej1·
Fable over 5.6 change my mind.
English
113
2
318
23.8K
Sean Murphy
Sean Murphy@AmbRealismX·
@elponick I have SOL doing pre-checks before pushing PR’s. If there is anything above low priority it gets sent back.
English
0
0
2
7
elponick
elponick@elponick·
@AmbRealismX planning in Sol High/Max + execution in Terra/Luna is beautiful. we do this - orchestrator drafts structural plans, specialized workers split implementation. keeps context clean. do you let workers check their own code against the plan or does Sol check everything?
English
1
0
2
11
Sean Murphy
Sean Murphy@AmbRealismX·
Anthropic gave everyone a nerfed Fable. OpenAI gave us a team of 3 each with up to 6 thinking levels. All three capable. Harness engineering & skill creation are more important than ever. You are the team leader and knowing who to put in the game and when will bring victories.
GIF
English
1
0
3
47
Sean Murphy
Sean Murphy@AmbRealismX·
@navigateny @DannyIO @DavidOndrej1 The 🔑 us in this graph. Think of 5.6 as a team. We have to put more thought into how we deploy the team. There are multiple points on the Sol line where Sol is not the best choice for the level of performance you get. Yes that and you will save tokens for better results.
Sean Murphy tweet media
English
1
0
1
13
navigator200
navigator200@navigateny·
Yes, he said this issue only happens if you run Sol on Ultra. That is not true based on my personal experience, and according to what others are posting on X they had a similar experience. When using Fable I feel it's just better. Smarter. A lot of experienced dev's are saying it's better than Sol, but that Sol is faster, and supposedly more token efficient and cheaper, but my issue with that last part is it seems to cook the 5 hour limits faster than anything I've seen.
English
1
0
0
28
Sean Murphy
Sean Murphy@AmbRealismX·
@simonw This graph should help Luna Ultra gives slightly more than Sol medium cheaper and faster. Look at them like a team. That’s the big difference. Anthropic gave us one, OpenAI gave us three with 6 thinking levels each. That is the code to crack - which team do you field?
Sean Murphy tweet media
English
0
0
1
559
Simon Willison
Simon Willison@simonw·
One of the most confusing aspects of GPT-5.6 is figuring out which model to use at which reasoning effort - sounds like Sol on Medium might be a good new default for coding work, if it's an upgrade from my previous favorite 5.5 xhigh
pash@pashmerepat

FYI: 5.6 Sol medium is better than 5.5 xhigh. As you go higher in reasoning levels on Sol, you will get insane levels of performance, but can burn through limits much more rapidly. We’re working on communicating this better!

English
78
20
875
96.2K
Sean Murphy
Sean Murphy@AmbRealismX·
@elponick I’m running it in Pi. Sol Xhigh or Max for the plan. Sol High orchestrating to Terra and Luna agents for implentation. Sol high checks the work before pushing PR’s per phase.
English
1
0
2
15
elponick
elponick@elponick·
@AmbRealismX curious about your actual harness setup here. are you running each tier as a separate agent with its own context window, or sharing state across all 3? we run 5 agents and the handoff layer is where everything breaks. one agent's output poisons the next one's context
English
1
0
2
25
Sean Murphy
Sean Murphy@AmbRealismX·
@navigateny @DannyIO @DavidOndrej1 Incorrect? Listen I burned $5800 worth of usage on the $200 Anthropic plan with Fable. It’s a great model. To say that it’s not even close is to say “I have not figured out how to prompt it”. Facts.
English
1
0
0
11
navigator200
navigator200@navigateny·
@DannyIO @AmbRealismX @DavidOndrej1 Incorrect. I ran it on high for 35 mins and it burned my entire 5 hour limit. I'm seeing dozens of people report the same thing. Even people on the $200 plan. Fable is the best coding model ever released, and it's not close.
English
1
0
0
22
Sean Murphy
Sean Murphy@AmbRealismX·
Curious why you would use Ultra for Planning? It has the ability to spawn a TON of subagents all using Ultra. Look at the delta in intelligence between High, X-High and Ultra it’s not that big of a gap. I think Ultra is for huge refactor jobs (rust conversions etc.) or really complex research tasks but for pure planning you’ll be spending under utilized tokens.
English
0
0
1
23
elCommendante
elCommendante@el_commendante·
@jumperz It is brutal, 5.6 sol ultra to make a plan -> Terra to execute. Way to go if you run out of resets, just hitting my first reset after 1 hour on one chat. But damn it it does the job
English
2
0
2
1.5K
JUMPERZ
JUMPERZ@jumperz·
GPT 5.6 early read > sol is the heavy hitter, but the usage burn is brutal > terra looks like the surprise winner so far > luna feels like the fast model people will actually use all day > the biggest upgrade is bug hunting / review > 5.6 is catching issues 5.5 X-High missed > repo-wide changes look much more confident > frontend taste still seems inconsistent > ultra sounds powerful, but not practical for casual use > best workflow: Sol for judgment, Terra/Luna for execution not ready to call it fable level, but this is clearly the first OpenAI drop in a while that feels like a real jump.
English
15
9
70
124.4K
Sean Murphy
Sean Murphy@AmbRealismX·
@NZT2048 Nice! I have been using it a ton, it is excellent at long running tasks and code reviews among other things. I hope you enjoy it! If you do please spread the word and give me a ⭐️ on GitHub if you feel it’s worthy! 🤙
English
0
0
1
39
LOGAN
LOGAN@NZT2048·
Anthropic sold the startup community that a marginally better model justifies 10x pricing. Insane marketing sorcery. Can’t even be mad tbh
LOGAN tweet media
English
2
1
2
2.8K
Sean Murphy
Sean Murphy@AmbRealismX·
@DavidOndrej1 Maybe it’s because they didn’t select you for the two month trial? Or maybe you are prompting it wrong? Selecting Ultra when high would do the trick? Tough to know without any further info from your vaguepost. I’m getting fantastic results ¯\_(ツ)_/¯
English
0
0
5
708
David Ondrej
David Ondrej@DavidOndrej1·
does OpenAI have half of the AI influencers on payroll? because I'm not impressed with GPT 5.6 at all...
English
155
3
396
36.1K
Sirui Zhang
Sirui Zhang@SZ_Sirui·
@ImSh4yy I consumed all of my 5 hour usage quota in 1 hour, reset and then consumed all the usage quota in 1 hour again. For the second time I was not even enabling fast mode or using ultra.
English
2
0
24
7.2K
Shayan
Shayan@ImSh4yy·
It seems like GPT 5.6 Sol is using my 5 hour codex limit much faster than GPT 5.5 despite it being more token efficient?
English
256
30
2.3K
297.9K

Keşfet

@NousResearch @kseniam0s @Merliiin__ @DavidOndrej1 @goof @thsottiaux @Youssofal_ @DannyIO