@benwoodward.bsky.social
Claude has a 'maximum conversation length' now???
i love Sonnet/Opus but some of the user-experience decisions Anthropic makes are ...puzzling.
i invest 20mins crafting a small dataset, and now i have to abandon the conversation because you're spreading your GPUs too thin?
i love Sonnet/Opus but some of the user-experience decisions Anthropic makes are ...puzzling.
i invest 20mins crafting a small dataset, and now i have to abandon the conversation because you're spreading your GPUs too thin?
June 22, 2025 at 12:22 AM
Claude has a 'maximum conversation length' now???
i love Sonnet/Opus but some of the user-experience decisions Anthropic makes are ...puzzling.
i invest 20mins crafting a small dataset, and now i have to abandon the conversation because you're spreading your GPUs too thin?
i love Sonnet/Opus but some of the user-experience decisions Anthropic makes are ...puzzling.
i invest 20mins crafting a small dataset, and now i have to abandon the conversation because you're spreading your GPUs too thin?
just got access to Claude voice mode, yessss
only speaks English, but the voices are very natural and realistic, UX is better than how I remember ChatGPTs working, although it still struggles with waiting it's turn to speak
only speaks English, but the voices are very natural and realistic, UX is better than how I remember ChatGPTs working, although it still struggles with waiting it's turn to speak
June 14, 2025 at 11:08 AM
just got access to Claude voice mode, yessss
only speaks English, but the voices are very natural and realistic, UX is better than how I remember ChatGPTs working, although it still struggles with waiting it's turn to speak
only speaks English, but the voices are very natural and realistic, UX is better than how I remember ChatGPTs working, although it still struggles with waiting it's turn to speak
it puzzles me that Anthropic has over a billion dollars in ARR, spends millions training models but adding a missing button in their mobile app is an insurmountable challenge
June 14, 2025 at 12:25 AM
it puzzles me that Anthropic has over a billion dollars in ARR, spends millions training models but adding a missing button in their mobile app is an insurmountable challenge
hate when this happens 😔
June 11, 2025 at 2:15 AM
hate when this happens 😔
this is such an amazing feature in Claude Code, i find it so much more intuitive to work with than Cursor, because you just tell it what you want, no configuration, or special buttons to press
June 11, 2025 at 12:12 AM
this is such an amazing feature in Claude Code, i find it so much more intuitive to work with than Cursor, because you just tell it what you want, no configuration, or special buttons to press
lmao at the way Gemini 2.5 Pro sometimes talks to itself
June 10, 2025 at 11:21 AM
lmao at the way Gemini 2.5 Pro sometimes talks to itself
hopefully they realise how much of a dumb idea this was and revert it in iOS 27, because I'm not upgrading until they do, because is just horrible
June 10, 2025 at 5:27 AM
hopefully they realise how much of a dumb idea this was and revert it in iOS 27, because I'm not upgrading until they do, because is just horrible
iOS 26 users trying to read their screens:
June 10, 2025 at 3:21 AM
iOS 26 users trying to read their screens:
tbh I don't want my buttons to be see through
June 10, 2025 at 3:16 AM
tbh I don't want my buttons to be see through
it only took millions of dollars training a 400 billion parameter large language model for me to understand my neovim config
May 23, 2025 at 7:33 AM
it only took millions of dollars training a 400 billion parameter large language model for me to understand my neovim config
Speak to Roam is quickly becoming the main way i'm taking notes recently.
super useful as a parent, don't get a lot of opportunity to sit down and take notes!
most of my big ideas hit me in flashes of inspiration, it's perfect for that. today i had a lot of ideas!
super useful as a parent, don't get a lot of opportunity to sit down and take notes!
most of my big ideas hit me in flashes of inspiration, it's perfect for that. today i had a lot of ideas!
May 9, 2025 at 2:41 AM
Speak to Roam is quickly becoming the main way i'm taking notes recently.
super useful as a parent, don't get a lot of opportunity to sit down and take notes!
most of my big ideas hit me in flashes of inspiration, it's perfect for that. today i had a lot of ideas!
super useful as a parent, don't get a lot of opportunity to sit down and take notes!
most of my big ideas hit me in flashes of inspiration, it's perfect for that. today i had a lot of ideas!
checking to see what's new in Arc is depressing now
April 29, 2025 at 3:16 AM
checking to see what's new in Arc is depressing now
I find this kind of content so unsettling, and it's so pervasive now.
members of the public are used as props for the content, filmed from a distance to capture their reaction to something.
members of the public are used as props for the content, filmed from a distance to capture their reaction to something.
April 21, 2025 at 10:30 AM
I find this kind of content so unsettling, and it's so pervasive now.
members of the public are used as props for the content, filmed from a distance to capture their reaction to something.
members of the public are used as props for the content, filmed from a distance to capture their reaction to something.
this is incredible. o4-mini-high (new OpenAI model) was able to solve this maze.
second image is the solution.
it reasoned about how to solve the problem, asked good clairfying questions and then write code to analyse the lines and create the red line as an overlay image
second image is the solution.
it reasoned about how to solve the problem, asked good clairfying questions and then write code to analyse the lines and create the red line as an overlay image
April 17, 2025 at 10:15 AM
this is incredible. o4-mini-high (new OpenAI model) was able to solve this maze.
second image is the solution.
it reasoned about how to solve the problem, asked good clairfying questions and then write code to analyse the lines and create the red line as an overlay image
second image is the solution.
it reasoned about how to solve the problem, asked good clairfying questions and then write code to analyse the lines and create the red line as an overlay image
just logged into Sora for the first time, and the realism of some of the images and videos is freaky
while remarkable, it does concern my that at some point we will have no reliable visual record of history
while remarkable, it does concern my that at some point we will have no reliable visual record of history
April 13, 2025 at 10:13 AM
just logged into Sora for the first time, and the realism of some of the images and videos is freaky
while remarkable, it does concern my that at some point we will have no reliable visual record of history
while remarkable, it does concern my that at some point we will have no reliable visual record of history
how am i supposed to know whether i'm using Gemini 2.5 or not Google?
April 11, 2025 at 2:02 AM
how am i supposed to know whether i'm using Gemini 2.5 or not Google?
whoah!
doesn't look exactly like Krashen, but it's pretty close. i take back what i said about 4o image gen. i guess it's just very hit and miss depending on what you're asking for.
doesn't look exactly like Krashen, but it's pretty close. i take back what i said about 4o image gen. i guess it's just very hit and miss depending on what you're asking for.
April 9, 2025 at 7:00 AM
whoah!
doesn't look exactly like Krashen, but it's pretty close. i take back what i said about 4o image gen. i guess it's just very hit and miss depending on what you're asking for.
doesn't look exactly like Krashen, but it's pretty close. i take back what i said about 4o image gen. i guess it's just very hit and miss depending on what you're asking for.
this better get more than 5 views! spent way too long editing this
April 9, 2025 at 5:18 AM
this better get more than 5 views! spent way too long editing this
Claude is so useful for creating quick visualisations for presentations and videos.
April 9, 2025 at 2:14 AM
Claude is so useful for creating quick visualisations for presentations and videos.
4o image gen. seems nerfed to me, cannot get it to do anything remotely close to the demo images.
just never quite follows my instructions however i prompt it.
just never quite follows my instructions however i prompt it.
April 9, 2025 at 12:03 AM
4o image gen. seems nerfed to me, cannot get it to do anything remotely close to the demo images.
just never quite follows my instructions however i prompt it.
just never quite follows my instructions however i prompt it.
been trying to get this video done for a month now (in between parenting and vibe coding), and i'm very close to the end
work in progress title: 'language is not lego', it's about grammar, and whether it actually exists in our heads, what the latest research says
1/3
work in progress title: 'language is not lego', it's about grammar, and whether it actually exists in our heads, what the latest research says
1/3
April 8, 2025 at 12:05 PM
been trying to get this video done for a month now (in between parenting and vibe coding), and i'm very close to the end
work in progress title: 'language is not lego', it's about grammar, and whether it actually exists in our heads, what the latest research says
1/3
work in progress title: 'language is not lego', it's about grammar, and whether it actually exists in our heads, what the latest research says
1/3
reactivated my ChatGPT subscription for 4o image gen. and Claude Sonnet is clearly more competent than all of OpenAI's models
doesn't matter which OpenAI model i use, it continually trips up on tasks that are straightforward for Sonnet
for example:
doesn't matter which OpenAI model i use, it continually trips up on tasks that are straightforward for Sonnet
for example:
April 7, 2025 at 1:10 AM
reactivated my ChatGPT subscription for 4o image gen. and Claude Sonnet is clearly more competent than all of OpenAI's models
doesn't matter which OpenAI model i use, it continually trips up on tasks that are straightforward for Sonnet
for example:
doesn't matter which OpenAI model i use, it continually trips up on tasks that are straightforward for Sonnet
for example:
if AI art looks terrible then why is a threat to real art?
April 1, 2025 at 10:23 PM
if AI art looks terrible then why is a threat to real art?
vibe coding in 2030
March 7, 2025 at 10:14 PM
vibe coding in 2030