@BenGrahamUK Where do I stand on a rainbow junction? I don’t. 😂
English
Jonathan Rudderham
2.8K posts
Jonathan Rudderham
@codeRunnerUK
Writing meets AI. Two hobbies combined. Oh, and a bit of running. And sometimes some coding. https://t.co/RWc3xfnLNG
Kent, UK Beigetreten Şubat 2009
110 Folgt184 Follower
This rainbow junction in Hounslow cost £48,174 of public money.
Some will say it promotes inclusion. Others will question the cost.
Where do you stand?
English
@piyush784066 Clearly not a serious question. You buy the one you can afford. The “coding” I do involves AI/LLM and neither of those two would do. Doesn’t mean the Neo isn’t ideal for some, especially those on a tight budget.
English
I stopped using Grok when I realised it had been corrupted to appease governments. The last straw was when I wanted it to generate a video of a scene from my story - the heroine sitting at the foot of a tree in a forest. I gave it no prompt, just the image.
The image was AI-generated. No real person involved. I had to fight tooth and nail to get Grok to admit that it won’t generate photorealistic humans in case the human looks like one of the 8 billion real people on Earth.
They’ve made it completely pointless.
English
Grok goes fully woke.
This is what $33 per month (billed annually) premium Grok gets you now.
No longer are you allowed to make artistic renderings of at-large crime suspects.
English
@tim_cook Given the millions of YouTube influencers extolling “adequate MacBook Neo” as “the second coming”, I’m not at all surprised. Why have a launch event when YouTube is full of people who are salesmen in the guise of reviewers?
English
@geekedout__ @LeakerApple And, yet, if I mention that it’s wrong for Apple to not “give a charger” at all when you spend almost £7,000 in Europe, I get piled on by Apple’s acolytes from the US for daring to mention it…
English
@LeakerApple it's a crime to give a 30W charger in a $1100 laptop 😭
English
The included 30W charger with my M4 MacBook Air is SLOOOOOOW.
English
@ravikiran_dev7 Some days, 128GB is not enough. And I’m not really a developer. I just do my own thing.
English
@jondelarroz I was kind of “meh…” and then … “Frank Castle! Yay, I’m watching this!” Would have been fantastic if DareDevil was to turn up given Matt Murdock’s previous cameo.
English
Why is Spider-Man: Brand New Day successful as a trailer?
Simple: the audience is starved for a white, male lead as a superhero after all of the MCU replacements.
English
@ivanfioravanti @UnslothAI Definitely excited for MLX training.
English
Love Unsloth Studio and I bet that when MLX training will be added it will be a super hit!
Well done @UnslothAI
English
I think people need to look at their use case, then do some testing on what they want it to do, and then decide. I don’t think anyone can decide for them, because we don’t all do the same thing. For me, 27B looks very promising, from 4bit to bf16. But I suspect, as I use them more, I’ll find bf16 works best for my use case (creative writing).
English
99% of people would be fine w/ qwen 27b and Hermes agent
The 1% can offload work to sota
First Squawk@FirstSquawk
GOOGLE HAS BEGUN TESTING A DEDICATED GEMINI APP FOR MAC TO COMPETE WITH CHATGPT AND CLAUDE, OFFERING FEATURES LIKE CONTENT GENERATION, WEB SEARCH, AND PERSONALIZATION.
English
@LLMJunky @TheAhmadOsman @0xSero @ivanfioravanti @ZenMagnets @NVIDIAAI It would take me more than 6 months to pay off the first one! 😂
English
I’ve been doing long context coherence tests on local LLM and there’s one thing I’ve learned - not all long context is the same. Some that claim to have 512K or 1M content window are just rubbish. Some models are doing well with 256K content window (the heavier Qwen3.5 models), but not all do.
English
Doing some experiments today with Opus 4.6's 1M context window.
Trying to push coding sessions deep into what I would consider the 'dumb zone' of SOTA models: >100K tokens.
The drop-off in quality is really noticeable. Dumber decisions, worse code, worse instruction-following.
Don't treat 1M context window any differently.
It's still 100K of smart, and 900K of dumb.
English
Reading posts on X, it feels like people using OpenClaw/HermesAgent are doing nothing more than jumping from LLM to LLM every few hours. Does anyone actually get any work done? 😂
English
@bindureddy Reading posts on X, it feels like people using OpenClaw/HermesAgent are doing nothing more than jumping from LLM to LLM every few hours. Does anyone actually get any work done? 😂
English
Open source continues to catch up with closed source
MiniMax 2.7 just dropped and will likely become the most used LLM on Claw in days 🚀
English
@Bitcoin_Teddy I don’t have student debt because we couldn’t afford for me to continue education after school. I got a job as soon as I could - before I even left school. I’ve never had job security in my whole life.
English
Older generations say “we all struggled in our 20s.”
No, you didn’t.
You didn’t pay $3200 for rent and $10 for eggs.
You didn’t graduate into $50K student debt and $0 job security.
Gen Z isn’t dramatic.
They’re drowning.
English
@4shpool I was not impressed with Nemotron-3-Super when I gave it my long context coherence test. It made me realise just how good the likes of Qwen-3.5 are (even though they’re not perfect).
English
He said
Quite literally
"Use a lower quant"
Q2 TRENCHERS REJOICE
Benjamin Marie@bnjmn_marie
Nemotron 3 Super GGUF evals: Q2 actually works. Model size goes from ~242 GB → ~53 GB, with only a tiny quality hit. Super impressive quantization robustness here, likely helped by NVFP4 pre-training, plus some very real Unsloth magic on top 🪄 Note: As usual, don't over interpret these results. +/- 10 points of relative error increase is not very meaningful. But above 10 points of increase, you can consider the model is bad.
English
@4shpool In my long context coherence test, Qwen 3.5 - 27B dense beat 35b-a3b hands down. For MoE, the 122B model was pretty much on par with the 27B dense model. I’ve not tested shorter context so YMMV.
English
@TommyFalkowski I found it to be hit-and-miss on my long context coherence test. Now that I've started testing non-Qwen3.5 models, it's showing how good Qwen3.5 is, but it's definitely a "your mileage may vary" model at 35B-A3B long context.
English
qwen3.5 35B-3A can actually do stuff. It's so weird and awesome to see a local model do useful things. Even with a pretty full context window it still seems usable.
English
@big_duca When I first heard "vibe coding", I had to look it up because I had no clue what it was. I code things that I want, not necessarily what other people want. They can code their own stuff. 🤣
English
@LottoLabs There seems to be quite a few in the arcee family, but I keep seeing "max_position_embeddings": 131072 so I think I'll leave this for a future test, not my long-context coherence test.
English