napier

Formerly, parts of the formerly free website formerly known as Twitter.

maybe he does. but which one is in prison?

The pricing seems... optimistic? It's not immediately clear to me what this offers that Google Lens/Gemini, Chat Gippity, or Claudius doesn't already do for free or zero marginal cost. Also specialised apps with basically fine-enough free tiers like PictureThis. Look for and focus on developing a unique angle -- and offer more for the money (or ask for a lot less from users). It's not facetious to say someone could replicate the current core features (a few prompted API calls and a wrapper) in the time it takes to boil a kettle. What's your moat?

Yeah, me. Even without other applications running in the background and without any models loaded, the new 0.3 UI is stuttering and running like a couch-locked crusty after too many edibles on my Macbook Air 2021, 16GB. When I finally get even a 4B model loaded, inference is glacially slow. The previous versions worked just fine (they're still available for download).

I what you're working on as a start! Reminds me a little of ofhttps://github.com/socketteer/loom. Composing multiple models in a guided / visually programable node flow of inputs and responses is a logical next step on from linear chatbots. The UI is feels a little raw at the moment -- zooming and scrolling / panning the canvas is necessary to keep track, which shouldn't change the visibility or zoom the lefthand pane as it currently does.

Would be nice to see a statement regarding user data security and privacy and how that's technically implemented.

You don't need to type on a hard surface (but anyway, most people tap away on their phone screens for hours in a day). You can type on an octopus if you want. Anything can be a projection surface keyboard with enough inspiration.

I suspect it's less about the music and more about the finger tracking, but to the extent it's about the music - billions of people have smartphones with cameras that would work with this code, but would unlikely be able to afford piano lessons or convince / inspire anyone to offer to purchase them an instrument without first demonstrating aptitude in some area. Definitely not intended at a dig at you and hope you manage to hook Mayuresh up with a keyboard instrument and a job!

this crops up in my feed every now and then and it has vastly superior perf vs. ØAI’s ChatGPT iOS app or anything else I’ve found. truly outstanding. are you planning on developing it further and/or monetizing it?

Don’t make users register/login before at least an introductory landing page flow. There’s currently no indication of what exactly I’m signing up for. Ideally also allow immediate usage for a reasonable number of minutes before splash prompting for the user to create an account so their progress will be saved, etc.

Their policy is to never talk about their content policy and to forbid their models from explaining when asked.

Fractal balkanization of general public attention context windows via social media simclusters and engagement maxxing feed algos.

Does it preserve readable (ad stripped) versions of pages I collect from falling to link rot, and does it automatically convert everything into content type adapative embeddings for citable chat querying with GPT-4 / user’s own personal LLM?

A gift for soon to be many multiples of the currently estimated 2 million rats in NYC.

There’s no evidence parachutes work.

Training on copyright data is arguably considered fair use in quite a few jurisdictions to various extents and levels of precedent, and entirely legal for entities based in Japan.

They almost certainly were. But the API only offers two default choices of GPT-4 (unless one has been anointed with exalted 32k access):

1. gpt-4-default which has been progressively nerfed with continuous no notification zero changelog zero transparency updates .

2. gpt-4-0314 which is a frozen checkpoint from shortly after public launch and is still great but not quite as good as the very original, or as flexible as the fridged based model. Fine. However it’s currently due to “no longer be supported” i.e. retired on June 14th.

It’s kind of a challenge to commit to building on murky quicksand foundations of an API product that changes drastically (but ineffably for the worse) without warning like the default accessible version does, and soon it looks like there won’t be a stable un-lobotomized alternative.

I would gladly pay triple digits a month for exactly that.

Ultra low contrast top header is hard to read. Red buttons mean stop. Ultra flat generic look. Invest in a few hours with a designer.

No pricing or byo API key option makes me more hesitant to commit. Intro video length unclear as YT controls disabled on embed, and can’t adjust playback speed. Differentiation from every other chat-with-your-docs option out there is unclear.

But congratulations on shipping a launch. I’m sure you can improve in tight iterations.

I’d like to see a model with the effluent of the internet intelligently filtered from the pretraining data by LLM and human curation, and much more effort to include digitised archival sources and the entirety of books and high quality media transcripts. I imagine it would yield far better baseline quality outputs with much less than current “requirements” for (over)correction with ultimately disastrous RLHF masking.

GPT-3 lineage model quality drastically declined over time from the initial launch of each too, with no transparency offered or critical feedback acknowledged by the lobotomy teams. It’s a shame to see it happening again and again.

No, not just you. It’s frequently abysmal nowadays. Tragic. And no version change-log or any other info from OAI about what they’ve done to it (probably GPU sparing optimized distillation + overly aggressive PR safety satisficing RLHF) and an advertised weeks long support query lag (I’ve yet to receive a response).

Municipal water fluoridation is problematic for a range of reasons but as a toothpaste and mouthwash it has merit and it actually does demonstrate direct bacterial growth arrest effects in addition to indirect substrate permissiveness inhibition via mineralisation.

“Fluoride acts to enhance membrane permeabilities to protons and compromises the functioning of F-ATPases in exporting protons, thereby inducing cytoplasmic acidification and acid inhibition of glycolytic enzymes.Basically, fluoride acts to reduce the acid tolerance of the bacteria. It is most effective at acid pH values. In the acidic conditions of cariogenic plaque, fluoride at levels as low as 0.1 mM can cause complete arrest of glycolysis by intact cells ofStreptococcus mutans.Overall, the anticaries actions of fluoride appear to be complex, involving effects both on bacteria and on mineral phases. The antibacterial actions of fluoride appear themselves to be complex but to be dominated by weak-acid effects.” https://cdnsciencepub.com/doi/10.1139/m95-133#:~:text=Basica....

I have the same question. It’s not clear what the usability differentiation is here. Why pay laughable amounts per question for GPT wrapper saas one trick ponies requiring email and credit card details instead of pulling one of hundreds of packages down from github (upon the backends of which these saas are built) that do exactly the same thing in exactly the same way, often with better functionality and UX for free with more active development and community support. The pricing of this as tends to be the case with these is err.. optimistic.

Is it possible to get access to the base model (ie no RLHF) via Azure? Or even the version that’s part of Goodbing?

What a clumsy half assed hack attempt at a hatchet job.

It’s an open and shut case.

Misleading paper title and misleadingly presented research findings. Models don't "degrade" over time; their timeliness and relevance remains frozen pursuant to fine-tuning updates while contemporaneous attentional demands upon them by users drift. They don't wear out like a pair of shoes.

very different initial approaches to training dataset curation, to SD's detriment.

Still waiting

Very much allegedly. They’re good, but not great.