[HN Gopher] Bard coding update with Colab exporting
___________________________________________________________________
Bard coding update with Colab exporting
Author : alphabetting
Score : 129 points
Date : 2023-04-21 13:45 UTC (9 hours ago)
(HTM) web link (blog.google)
(TXT) w3m dump (blog.google)
| renewiltord wrote:
| It actually got much faster and much better at code. Impressive!
| srameshc wrote:
| I am using it right now to debug some of my Go and SQL code and
| it's works great.
| butz wrote:
| Could someone ask Bard how to bypass its region-lock?
| bl4ck_goku wrote:
| I used a VPN.
| londons_explore wrote:
| Just did a comparison of Bard vs GPT-4 with a coding question I
| came up with:
|
| > I have a 3 dimensional black bitmap (voxel map) with a white
| straight line drawn in it. The 3d bitmap is saved as a bunch of
| tiff files by layer eg. layer_347.tif Write me python code to
| load this data and return the x,y,z location of each end of the
| line. The bitmap is too big to load all of it into RAM at once.
| Use a binary search to avoid loading unnecessary files.
|
| Bard came up with a solution which was close but didn't follow
| the instructions (it just loaded all the data). It also assumed
| layer_9 sorted before layer_10.
|
| GPT-4 came up with a solution which followed the instructions,
| but had a buggy binary search implementation, and a suboptimal
| O(n).
|
| The GPT-4 solution was closer to what I was after, but there
| wasn't much in it. Neither implementation handled edge cases like
| missing files or had tests or anything.
| belltaco wrote:
| > Neither implementation handled edge cases like missing files
| or had tests or anything.
|
| Did you tell them to do those things after they came up with
| the solution?
| ThalesX wrote:
| I'm sorry for the confusion. Here's a version that handles
| edge cases:
|
| ```
|
| // Basically still not handling edge cases
|
| ```
| coding123 wrote:
| [flagged]
| zb3 wrote:
| Also this nonsense geoblock that still exists. I've already
| joined waitlist via Tor and then I was granted access. Yet when
| I visit bard.google.com without Tor I'm still informed that
| bard isn't supported in my country.
| benatkin wrote:
| I like not being able to sign up because I'm under 18.
| Preferably in hexadecimal. https://www.reddit.com/r/ChatGPT/c
| omments/11ydh2y/im_trying_...
| lyu07282 wrote:
| [flagged]
| JLCarveth wrote:
| I have full access to Bing Chat in Canada, but not to bard.
| Disappointing.
| quadrature wrote:
| I think this is fine, the problem is that they aren't calling
| it a beta/preview which would temper people's expectations.
| thebestmoshe wrote:
| > Bard is still an early experiment, and may sometimes
| provide inaccurate, misleading or false information while
| presenting it confidently. When it comes to coding, Bard may
| give you working code that doesn't produce the expected
| output, or provide you with code that is not optimal or
| incomplete. Always double-check Bard's responses and
| carefully test and review code for errors, bugs and
| vulnerabilities before relying on it.
| danpalmer wrote:
| It's labelled as "Experiment" in the titlebar, it says that
| it has limitations and that feedback will help improve, and
| it says that it may display inaccurate information. I think
| that covers it all pretty well.
| quadrature wrote:
| nice, missed that on the bard site.
| heliophobicdude wrote:
| Something I am just now noticing is that Barb is able to view the
| contents of a pdf that I provided a link to.
|
| Could be making it up but need to test further...
| karussell wrote:
| At least Google seems to respect open source unlike OpenAI: "If
| Bard quotes at length from an existing open source project, it
| will cite the source."
| indymike wrote:
| So... does that imply that it simply fails to cite where it's
| quoting from closed source projects?
| benatkin wrote:
| That's also likely a better way to exploit open source.
|
| Microsoft and OpenAI got a lot right but also a lot wrong, I
| think. This could have held Copilot back, not making proper
| composite projects.
| pnathan wrote:
| I'm experimenting with using Bard to generate Prolog code. I have
| this idea that if I can synthesize Prolog and a LLM, I can do
| some very clever planning work.
| binkHN wrote:
| Show me the API.
| UncleOxidant wrote:
| I just gave Bard a try. It's still nowhere near as good at coding
| as chatGPT (and GPT4 specifically). I asked it to write a Python
| program to solve the Traveling Salesman Problem using ACO. The
| code GPT4 gave me for this prompt runs out of the box which is
| pretty impressive. The Bard code, on the other hand, has errors
| like forgetting to pass an argument to a function, or calling a
| method on a class that hasn't been defined on that class. When I
| point out the error to Bard it presents code that it says solves
| the problem, but often doesn't. In some cases when it fixes a
| problem in one area it reintroduces a previously fixed bug
| elsewhere.
|
| Also, chatGPT always goes the extra mile and includes
| instructions on compiling/running the code it provides (including
| sample data). Bard does not.
|
| Bard still has a ways to go to catch up.
| tyfon wrote:
| I've found that GPT4 does very well for problems that have
| examples on the web, but I have yet to make it display a 3d
| cube using opengl (any version) in linux using c++ or python.
|
| So it depends heavily on the task and how much it has stored
| from the training.
|
| Now if you ask them to do a teapot, it sometimes do it since
| there are so many examples of that, even if you often have to
| tinker a bit to get it to compile and run, at least it is a
| teapot and not some weird random geometrical object.
|
| I have yet to try it with bard, will test when I get home to a
| proper computer.
| amf12 wrote:
| > I just gave Bard a try. It's still nowhere near as good at
| coding as chatGPT (and GPT4 specifically)
|
| Because its a much smaller model IMO. Its like comparing a car
| with 2.0 engine, with a car with 3.0 engine. Also ChatGPT made
| a lot of errors in code just a few months ago. Google seems to
| be catching up.
| verdverm wrote:
| Worth noting that Google i/o is next month and many are
| expecting announcements related to Bard. There is that rumor
| floating around that they are training a very large (30T)
| model
| thewataccount wrote:
| The only source I can find is a reddit post from April
| fool's day, do you have a better one by chance?
|
| There's almost no way they make a 30T model that's useful
| at least? That's over 60TB of memory at 16bit, 15TB at 4
| bit. That's 15TB that have to go through either VRAM or TPU
| Memory (TPUv4 is 32GB I think?). Yes you could distribute
| it and chunk it, but that would be insanely slow/expensive
| to run for anything more then a research demo.
|
| GPT-4 is most likely 1~2T from what I've heard, and they've
| been struggling to run that profitably/quickly judging by
| the crazy pricing and 25messages/3hours if you pay for
| premium chatgpt.
|
| EDIT: The only way this is possible is if they've pulled
| off some crazy optimization, which is definitely possible,
| but a lot of the research from them such as their own
| Chinchilla scaling suggest smaller models that are more
| fully trained are better.
|
| According to their chinchilla paper they would need over
| 330TB of training data for that, and an absurd amount of
| training time.
| verdverm wrote:
| See my other comment:
| https://news.ycombinator.com/item?id=35657567
| politelemon wrote:
| I've been using it extensively and never once found ChatGPT to
| produce working code when the ask is just above basic. It might
| have to do with subject matter. It usually is close enough but
| my experience has been what you've found with Bard - missing
| arguments, phantom methods, etc.
|
| Haven't bothered with Bard at all.
| UncleOxidant wrote:
| did you try with GPT4? I was able to run one GPT4 query/day
| for free on poe.com
|
| Yeah, given the random nature of these LLMs it's possible
| that I won the lottery in that particular case.
| fhd2 wrote:
| In my experience, GPT-4 does get lots of textbook/common
| problems right in common languages on first try. But when I
| tried to solve a just mildly complex problem (finding a
| ring of cells in a Godot GridMap using GDScript), it
| produced nonsense: Solved the problem wrong, called
| functions that don't exist and so on. It did correct some
| of the mistakes after I pointed them out, but I couldn't
| get something working without manual editing. Then I asked
| it to optimise the solution and got nonsense again. For
| simple data wrangling in Python it worked a lot better.
|
| I have a suspicion there might be some... showmanship at
| work where all kinds of test problems and common tasks are
| made sure to work well.
|
| Also it generates code slower than I would type (if I know
| what to write), which makes me mad for some reason.
| fnordpiglet wrote:
| I think it has a lot to do with how common code examples
| in the language are. I bet if you tried with languages
| detailed here you will get a lot better results than more
| obscure languages (like GDscript)
|
| https://octoverse.github.com/2022/top-programming-
| languages
| busseio wrote:
| [dead]
| bg24 wrote:
| Tried it and love it. This is still not at par with ChatGPT 3.5.
| I use ChatGPT all the time, and will increase use of Bard from
| now.
|
| Good that Bard is aware of latest version of software (eg.
| NextJS). Helps when APIs of a software evolves quickly.
|
| I use a lot of Google products and really hope that they manage
| to nail it as at least #2 player in the space. As a company, they
| have not proven to come back and take the lead from behind.
| r3trohack3r wrote:
| GPT-4 has a lot of problems with recent versions of APIs.
| Working with Cloudflare workers was initially painful because
| of the dated code it would generate. The code would be valid,
| it just didn't work with the initial scaffolding wrangler
| generated.
|
| I've found copying and pasting chunks of the API context into
| the prompt solves 99% of it. I'm now able to get it to generate
| modern workers just fine. You can copy an example method
| signature, some docs, etc. and it picks up on that fairly
| reliably. I've even just copy-and-pasted an entire worker
| implementation (100+ LoC) at the beginning of a prompt and that
| worked.
| bg24 wrote:
| Would you mind sharing 1 example how you do it? Or any web
| link?
|
| I have spent a lot of time reading through prompt
| engineering, but coding examples to adapt to latest versions
| are rare to find.
| r3trohack3r wrote:
| I don't bother much with prompt engineering. I just treat
| ChatGPT as a pair programmer who can't see my editor. I
| tell it where files are, what I'm using, copy and paste
| relevant code in, and it just works.
|
| The big "breakthrough" for me was unlearning all of my
| search engine habits of thinking I need to be precise and
| talk to it like a machine. I use compliments like "that's
| great" and it seems to reinforce good answers well enough.
|
| Just treat it like a human on the other side of the screen
| and adjust as necessary.
|
| Also pay the $20 - GPT-3.5 is not good, GPT-4 is junior
| engineer good. Most of my time is now spent describing
| problems to it and code reviewing what it does.
|
| Edit: I do use prompt engineering when I'm building AI
| enabled products to help keep things on the rails. But for
| my personal use of ChatGPT I just let 'er rip.
| tethys wrote:
| They gave an example of this in the launch live stream,
| about 12 minutes in:
| https://www.youtube.com/live/outcGtbnMuQ
| shmoogy wrote:
| Have you tried phind.com ? It scrapes sites so it can get
| additional context
| codemac wrote:
| I've found that the bugs that GPT-4 and GPT-3.5 make are
| actually... somehow more nefarious because it understands
| enough to make very complex bugs.
|
| The complexity of the bugs you get are usually 10x as complex
| as the code's main logic you wrote. So you need to write very
| simple code.
|
| Unfortunately, these GPT models are very complex. The bugs they
| produce will take an order of magnitude better model to fix.
| javier_e06 wrote:
| I wouldn't touch anything that Google(The eyes and ears of
| advertisers) offers, with a 10 foot pole. Specially if is work
| related. Chat GPT software development answers get better with
| warm up questions. I don't shoot questions point blank but do a
| 3-4 related questions first.
| neel8986 wrote:
| You do realize that ChatGPT is owned partially by another ad
| gint Microsoft who is also hell bent on showing ads in start
| menu
| javier_e06 wrote:
| Yeah, I wouldn't touch MS Windows or Bing either. Is Duckduck
| Go and Linux for me. If ChatGPT starts charging for basic
| service, I might pay.
| sidcool wrote:
| It's not available in my country yet.
| eigenvalue wrote:
| I don't really see why anyone would use this unless they've been
| in a coma for the past few months and never heard of ChatGPT.
| r0m4n0 wrote:
| Why does competition exist ever? I think there are trade offs
| for everyone to consider before they blindly assume one will be
| the better option for every use case forever
| jeffbee wrote:
| I try to glance at both, but because I do that I know that
| ChatGPT has severe availability problems, whereas Bard so far
| has not failed for me. And Bard is much faster even though
| superficially it can seem slower because ChatGPT blurts out the
| first word more quickly.
| danpalmer wrote:
| Well right now it's just another website you have to go to with
| intent to use it like ChatGPT, but if Google built it into
| search it would get a ton of organic use from people doing
| things like googling error messages.
|
| I'm not sure if this is what is planned of course, but that
| would be far more widely used than an otherwise siloed website
| you have to visit specifically.
|
| Also remember that most people have at least never used
| ChatGPT, and most probably still haven't heard of it. We live
| in a tech bubble - none of my family had heard of it a few
| weeks ago. They all use Google search on a daily basis.
| rexbee wrote:
| Bard seems to use current data whereas Chatgpt stops at 2021.
| Bard is also significantly faster at returning responses.
| godzillabrennus wrote:
| Being trained on more up to date data and being faster are
| irrelevant when returning responses that aren't as useful.
| Google has catching up to do.
| quadrature wrote:
| It's good to have more competition in the space. An AI space
| race between Google and OpenAI could be really interesting.
| ShamelessC wrote:
| Competitions have winners and losers.
| benatkin wrote:
| Think of ChatGPT as Breaking Bad and Bard as Better Call Saul.
| mrits wrote:
| I've asked bard some coding questions around winapi that I've
| asked ChatGPT the last few weeks. So far not only does it give
| incorrect answers, it seems super confident in telling me the
| wrong thing.
|
| E.g, does SendInput send to a background window? It told me not
| only does it but it is the only way to do it and then gave me
| examples where it got a handle to a window but then didn't even
| use it.
|
| A few months ago this would be super impressive but after using
| ChatGPT there is no reason that I've found why I'd want to use an
| inferior AI.
| kkkkkkk wrote:
| FWIW your example seems to work for me.
|
| https://i.imgur.com/1sWc2WR.png
| mrits wrote:
| https://imgur.com/a/qmjVPJ5
|
| Same question with a different answer.
|
| This is interesting. This version of the question is better
| than the original answer I got but still not as good as your
| answer.
| impulser_ wrote:
| It's pretty cool that it will cite open source if it quotes code
| from an open source project, unlike ChatGPT and Copilot.
| andai wrote:
| To comply with GPL, it would have to automatically re-license
| your project as GPL too, right?
| vineyardmike wrote:
| If _you_ use GPL code, _you_ may have to relicense _your_
| project.
|
| _Bard_ can share with you any code that it is licensed to
| distribute (which includes GPL).
| procinct wrote:
| But copilot actually outputs code into your project, do you
| really want citations there? I prefer to just enable the
| setting for it to not reproduce code verbatim from other
| projects.
| [deleted]
| weakwire wrote:
| It's really fast! That is very important. Also it provides 3
| variation drafts. I see this as a winner
| cush wrote:
| The Copilot VSCode extension does 10 drafts
| londons_explore wrote:
| Both can be made to give more drafts by just asking the same
| question again...
|
| I don't see more drafts as a real differentiator.
| weakwire wrote:
| it spits out 3 drafts in 1 go. without the need to wait. It
| is simply way faster than GPT4
| verdverm wrote:
| I wonder how much of that has to do with TPUv4 vs the
| hardware used for GPT4?
|
| Google has invested in custom AI hardware for some time
| now and does not run their workloads on nvidia cards
| londons_explore wrote:
| Neural networks are really parallelizable. If I scale up
| my AI service to handle double the number of users by
| buying double the number of GPU's, it is theoretically
| possible to _also_ serve each user in half the time.
|
| To do so, you need to split the matrix multiplies across
| the new machines. You also need more inter-machine
| network bandwidth, but with GPT-3 that works out to 48
| kilobytes per token predicted collected from every
| processing node and given to every processing node. Even
| if Bard is 100x as big, that is still very doable within
| datacenter scale networking.
|
| However, OpenAI doesn't seem to have done this - I
| suspect an individual request is simply routed to one of
| n machine clusters. As they scale up, they are just
| increasing n, which doesn't give any latency benefit for
| individual requests.
| verdverm wrote:
| Yup, the TPUv4 pod is highly optimized
|
| They are claiming to be the first to achieve >50%
| saturation during training. Pretty sure I recall
| Midjourney is using TPUv4 pods too
|
| https://cloud.google.com/blog/products/ai-machine-
| learning/g...
|
| https://cloud.google.com/tpu/docs/system-architecture-
| tpu-vm
| lavasalesman wrote:
| UX is a feature
| ______ wrote:
| Doesn't work for this simple prompt -- maybe it's not fully
| rolled out yet?
|
| > Can you write a <HTML> file that POSTs a form with firstName,
| lastName to some other URL?
|
| > I'm a language model and don't have the capacity to help with
| that.
| gmm1990 wrote:
| Reworded the prompt and it seemed to work, I entered yours and
| it didn't work but the below did: how would you send a post
| html request with firstName, lastName as inputs to a url
|
| To send a POST HTML request with firstName and lastName as
| inputs to a URL, you can use the following steps:
|
| 1. Create an HTML form with two input fields, one for firstName
| and one for lastName. 2. Set the method of the form to "POST".
| 3. Set the action of the form to the URL you want to send the
| request to. 4. Submit the form.
|
| The following is an example of an HTML form that you can use to
| send a POST request with firstName and lastName as inputs:
|
| ```html <form method="POST"
| action="https://example.com/api/v1/users"> <input type="text"
| name="firstName" /> <input type="text" name="lastName" />
| <input type="submit" value="Submit" /> </form> ```
|
| When you submit this form, the server will receive the
| firstName and lastName values as POST request parameters. You
| can then use these values to perform whatever actions you need
| to do on the server.
| oh_sigh wrote:
| It's probably just confused that you're asking it to "write a
| file" and not tell you the contents of what the file should be
___________________________________________________________________
(page generated 2023-04-21 23:02 UTC)