HNHacker News
TopNewBestAskShowJobs

tikkun

2,864 karma · joined May 19, 2022

submissionscomments
tikkun··on Meta's new AI image generator was trained on 1.1B Instagram and FB photos
I tried it now.

My experience:

Took 4 minutes to log in and do one generation. (Login to FB, then it took me through a process to merge accounts with Meta, which didn't sound good, so I restarted with 'sign in via email' which ended up doing the same thing anyway, I think. Then I was logged in, did the generation.)

My at a glance is that it's:

For image quality

1. Midjourney

2. Dall e 3

3. SDXL and this

For overall ease of use and convenience

1. Dall e 3

2. Midjourney

Of course, this is all biased personal opinion, and YMMV.

tikkun··on Gemini AI
My general conclusion: Gemini Ultra > GPT-4 > Gemini Pro

See Table 2 and Table 7 https://storage.googleapis.com/deepmind-media/gemini/gemini_... (I think they're comparing against original GPT-4 rather than GPT-4-Turbo, but it's not entirely clear)

What they've released today: Gemini Pro is in Bard today. Gemini Pro will be coming to API soon (Dec 13?). Gemini Ultra will be available via Bard and API "early next year"

Therefore, as of Dec 6 2023:

SOTA API = GPT-4, still.

SOTA Chat assistant = ChatGPT Plus, still, for everything except video, where Bard has capabilities . ChatGPT plus is closely followed by Claude. (But, I tried asking Bard a question about a youtube video today, and it told me "I'm sorry, but I'm unable to access this YouTube content. This is possible for a number of reasons, but the most common are: the content isn't a valid YouTube link, potentially unsafe content, or the content does not have a captions file that I can read.")

SOTA API after Gemini Ultra is out in ~Q1 2024 = Gemini Ultra, if OpenAI/Anthropic haven't released a new model by then

SOTA Chat assistant after Bard Advanced is out in ~Q1 2024 = Bard Advanced, probably, assuming that OpenAI/Anthropic haven't released new models by then

tikkun··on Gemini AI
To add to my comment above: Google DeepMind put out 16 videos about Gemini today, the total watch time at 1x speed is about 45 mins. I've now watched them all (at >1x speed).

In my opinion, the best ones are:

* https://www.youtube.com/watch?v=UIZAiXYceBI - variety of video/sight capabilities

* https://www.youtube.com/watch?v=JPwU1FNhMOA - understanding direction of light and plants

* https://www.youtube.com/watch?v=D64QD7Swr3s - multimodal understanding of audio

* https://www.youtube.com/watch?v=v5tRc_5-8G4 - helping a user with complex requests and showing some of the 'thinking' it is doing about what context it does/doesn't have

* https://www.youtube.com/watch?v=sPiOP_CB54A - assessing the relevance of scientific papers and then extracting data from the papers

My current context: API user of OpenAI, regular user of ChatGPT Plus (GPT-4-Turbo, Dall E 3, and GPT-4V), occasional user of Claude Pro (much less since GPT-4-Turbo with longer context length), paying user of Midjourney.

Gemini Pro is available starting today in Bard. It's not clear to me how many of the super impressive results are from Ultra vs Pro.

Overall conclusion: Gemini Ultra looks very impressive. But - the timing is disappointing: Gemini Ultra looks like it won't be widely available until ~Feb/March 2024, or possibly later.

> As part of this process, we’ll make Gemini Ultra available to select customers, developers, partners and safety and responsibility experts for early experimentation and feedback before rolling it out to developers and enterprise customers early next year.

> Early next year, we’ll also launch Bard Advanced, a new, cutting-edge AI experience that gives you access to our best models and capabilities, starting with Gemini Ultra.

I hope that there will be a product available sooner than that without a crazy waitlist for both Bard Advanced, and Gemini Ultra API. Also fingers crossed that they have good data privacy for API usage, like OpenAI does (i.e. data isn't used to train their models when it's via API/playground requests).

tikkun··on Gemini AI
Yes, and presumably: more data, more compute, better pre-training and post-training methods, leading to better capabilities.
tikkun··on Gemini AI
One observation: Sundar's comments in the main video seem like he's trying to communicate "we've been doing this ai stuff since you (other AI companies) were little babies" - to me this comes off kind of badly, like it's trying too hard to emphasize how long they've been doing AI (which is a weird look when the currently publicly available SOTA model is made by OpenAI, not Google). A better look would simply be to show instead of tell.

In contrast to the main video, this video that is further down the page is really impressive and really does show - the 'which cup is the ball in is particularly cool': https://www.youtube.com/watch?v=UIZAiXYceBI.

Other key info: "Integrate Gemini models into your applications with Google AI Studio and Google Cloud Vertex AI. Available December 13th." (Unclear if all 3 models are available then, hopefully they are, and hopefully it's more like OpenAI with many people getting access, rather than Claude's API with few customers getting access)

tikkun··on LLM Visualization
Oh: https://news.ycombinator.com/item?id=38511659
tikkun··on LLM Visualization
What happened to this thread? When I saw it before it had 700+ upvotes.
tikkun··on Music-Map – Find Similar Music
This is the best one. Such a cool site.
tikkun··on Ask HN: How can you export iPhone Messages (e.g. to a CSV)?
Search github there are a few utilities on there for this
tikkun··on Create a cute avatar from your photo. Powered by DALL-E 3
Not my product, but looks cool.
tikkun··on Together AI raises a $102.5M Series A
I've explored most of the inference as a service platforms, theirs is particularly nice. (Clear/simple UI, good pricing, good speed.)
tikkun··on Undercover in Saudi Arabia's secretive program to keep the world burning oil
> It's also spending $270 trillion by 2030 on low-carbon energy initiatives

Article says $270 billion, not trillion

tikkun··on Apple unveils the top books of 2023 and a new Year in Review experience
https://www.apple.com/newsroom/2023/11/apple-shares-the-most...

https://www.apple.com/newsroom/2023/11/apple-unveils-the-top...

OP's post was missing the links

tikkun··on Silicon Valley's Laundry-App Race (2014)
Laundry Buddy is the real winner here.
tikkun··on List of AI safety challenges and meta-level ideas for addressing some
Let me know what “challenges of today” you’d add to the list. I put this together recently and wrote it up and shared it with a few friends.
tikkun··on Launched Playtoob, a simple YouTube companion, because my 5yo daughter loves it
Related: I recently learned that in the youtube kids app the parent can sign in and select only pre-approved accounts, and can disable search and recommendations.
tikkun··on Ask HN: I've spent $1M+ building a fitness app. Now what?
What's your financial standing? Are you doing this for fun and you already have enough assets to live on, or does this need to work financially?

If it needs to work, I think pausing development and focusing only on marketing, as you say, seems like a good path. And I'd try and do "high fidelity" marketing rather than low fidelity - e.g. approach people in person at the cafe at a gym or after their workout, rather than sending ads or emails into the void where you don't know what lands.

I'll note that it's kind of wild to have invested $1mm without a seemingly clear plan for distribution, but, it seems that you're taking steps to address that now at least.

tikkun··on Startup Boards: A Field Guide to Building an Effective Board of Directors
And then also:

Board Member Orientation: The Concise and Complete Guide to Nonprofit Board Service

https://www.amazon.com/Board-Member-Orientation-Complete-Non...

tikkun··on Inflection-2: the next step up
Regular reminder that most open source LLM benchmarks are not very useful (in the sense that they don't represent day to day ai chatbot usage and what users care about). If you haven't looked through the datasets to see what they actually contain, I'd encourage you to do so. [1] I think we're just in a strange suboptimal schelling point of sorts, where people report their scores on those benchmarks because they think other people care about those sort of benchmarks, and therefore those benchmarks are the ones that people expect and care about.

And to recap their statement about it being second most powerful, it's based on MMLU scores, which IMO is a non-useful comparison. (Also, doesn't test against GPT-4-Turbo or Claude-long-2.1)

What they're saying is that Inflection-2 ranks #2 relative to other models including GPT-4, Claude-2, PaLM 2, Grok-1, and Llama 2 70b, specifically on MMLU scores.

This model could be great, but that'll be determined by "do day to day users, both free and paying, prefer it over Claude 2 and GPT-4-Turbo" - not MMLU scores.

[1]: https://huggingface.co/datasets/lukaemon/mmlu/viewer/abstrac...

tikkun··on "Et Tu, Ilya?"
I do hope that Ilya gets good support! I hope the same for Sam, all other employees. I’m worried that Ilya will get a lot of hate. (My allegiance is with Sam.) But I support general kindness and I feel sadness for all involved.
tikkun··on "Et Tu, Ilya?"
My hope for the situation is that we can give as much compassion as possible to all involved.

My hope is that the people around Ilya give him compassion and love. He will be experiencing sadness, regret, and grief. I hope he gets lots of good emotional support.

Same for Sam, same for all employees.

My hope for Ilya is that once the dust settles, instead of getting people scared about AI safety by talking to them, he gives people the experience he had. Give the training data, not the output. You can't reason someone into or out of a position that they didn't reason themselves into. People don't get into opinions from a place of reasoning but from a place of emotion and experience. To help someone shift their opinion, we need to give them experience.

There are some interesting religious lenses on the situation. It certainly will remind people a bit of a crucifixion scenario where we may see Sam and Greg emerge three days later within a new company. It's kind of like a Phoenix from the ashes, like Aslan in Narnia, which of course is also based on the story of Christ. Humans like stories of scapegoats and people being attacked and then being defended. One of the things we've seen is a massive outpouring of support for Sam Altman on Twitter, particularly from so many random founders and people in tech that he helped whether they were nobodies or already successful. When someone gets attacked publicly in a way that people dislike or when it’s a person people really like and like more than the perceived attacker, then they have a lot of other people that then want to come and defend them.

I hope everyone involved realizes that debate doesn't accomplish much other than polarization and tribalization, and that people instead seek to make the other side feel understood. And give the other side the experiences that gave them their own preferences/feelings, rather than trying to convince them to feel the same way. And generally, help people emotionally regulate instead of going into debate or commiserating by sharing negative things about other parties involved.

(Don’t tweet while angry, instead, share your emotions with a friend, then tweet after you’re feeling neutral and peaceful!)

Overall, I hope people minimize attacks. I hope that people directly express their own emotions so that they don't come out in other ways cloaked in rationalizations. This will help people avoid derogatory comments towards people. Sharing specific facts and specific stories can be okay, but avoiding miscellaneous negative judgment in any direction. I hope that people from all directions can see that everyone involved is just doing what they felt was good, and there's a good chance that you would have done something similar if you had the same genetics and upbringing. I hope everyone involved realizes that debate doesn't accomplish much other than polarization and tribalization…

Reminder: Any attack or vitriol makes the person defend themselves, and defending themselves make them more of whatever they are defending themselves against.

There will be a lot of people experiencing varying degrees of grief in the startup and AI ecosystems in the coming week.

Ideally, everyone involved just goes in and gets a really good Swedish relaxation massage, is feeling super peaceful, helping all those around them feel equally peaceful, focusing on releasing and processing their emotions. I believe this moment is potentially a critical moment in history, and the more emotional regulation the people involved in the industry can do over the coming weeks, the better. I believe the future is path-dependent and depends on the present. Generally, we get more of what we have, so the more peace we can have in the present, the more peace we're likely to have in the future. I hope we can all maximize empathy and understanding for all the people involved and minimize polarization and tribalization. I hope people try to make others feel understood and focus on emotional regulation.

Emotional regulation and open-mindedness are vital when facing unexpected and influential incidents.

tikkun··on OpenAI staff threaten to quit unless board resigns
This situation will create the need to grieve loss for many involved.

I wrote some notes on how to support someone who is grieving. This is from a book called "Being There for Someone in Grief." Some of the following are quotes and some are paraphrased.

Do your own work, relax your expectations, be more curious than afraid. If you can do that, you can be a powerful healing force. People don't need us to pull their attention away from their own process to listen to our stories. Instead, they need us to give them the things they cannot get themselves: a safe container, our non-intrusive attention, and our faith in their ability to traverse this road.

When you or someone else is angry, or sad, feel and acknowledge your emotions or their emotions. Sit with them.

To help someone heal from grief, we need to have an open heart and the courage to resist our instinct to rescue them. When someone you care about is grieving, you might be shaken as well. The drama of it catches you; you might feel anxious. It brings up past losses and fears of yourself or fears of the future. We want to take our own pain away, so we try to take their pain away. We want to help the other person feel better, which is understandable but not helpful.

Avoid giving advice, talking too much, not listening generously, trying to fix, making demands, disappearing. Do see the other person without acting on the urge to do something. Do give them unconditional compassion free of projection and criticism. Do allow them to do what they need to do. Do listen to them if they need to talk without interruptions, without asking questions, without telling your own story. Do trust them that they don't need to be rescued; they just need your quiet, steady faith in their resilience.

Being there for someone in grief is mostly about how to be with them. There's not that much you can "do," but what can you do? Beauty is soothing, so bring fresh flowers, offer to take them somewhere in nature for a walk, send them a beautiful card, bring them a candle, water their flowers, plant a tree in honor and take a photo of it, take them there to see it, tell them a beautiful story about the thing that was lost from your memory, leave them a message to tell them “I’m thinking of you”. When you’re together with them in person, you can just say something like "I'm sorry that you're hurting," and then just kind of be there and be a loving presence. This is about how to be with someone for the grief message of a loss of a person. But all the same principles apply in any situation of grief, and there will be a lot of people experiencing varying degrees of grief in the startup and AI ecosystems in the coming week.

Who is grieving? Grieving is generally about loss. That loss can be many different kinds of things. OpenAI former and current team members, board members, investors, customers, supporters, fans, detractors, EA people, e/acc people, there’s lots of people that experienced some kind of loss in the past few days, and many of those will be grieving, whether they realize it or not. And particularly, grief for current and former OpenAI employees.

What are other emotional regulation strategies? Swedish massage, going for a run, doing deep breathing with five seconds in, a zero-second hold, five seconds out, going to sleep or having a nap, closing your eyes and visualizing parts of your body like heavy blocks of concrete or like upside-down balloons, and then visualize those balloons emptying themselves out, or if it's concrete, first it's concrete and then it's kind of liquefied concrete. Consider grabbing some friends, go for a run or exercise class together. Then if you discuss, keep it to emotions, don’t discuss theories and opinions until the emotions have been aired. If you work at OpenAI or a similar org, encourage your team members to move together, regulate together.

tikkun··on Ask HN: Resources to brush up from 'Intro to ML' to current LLMs/generative AI?
And perhaps things like: https://www.youtube.com/watch?v=VPRSBzXzavo and http://jalammar.github.io/illustrated-transformer/ and https://peterbloem.nl/blog/transformers

See also: https://news.ycombinator.com/item?id=35990118

tikkun··on Ask HN: Resources to brush up from 'Intro to ML' to current LLMs/generative AI?
I think you're asking for some of the things mentioned in the 'research science' section here - https://news.ycombinator.com/item?id=36195527 - is that right?
tikkun··on PSA: GPT4 (web) token limit down to 4k
It’s 128k tokens of input and 4096 tokens of output (gpt-4-turbo), though in chatgpt it’ll be different because it uses progressive summarization for long conversations. ChatGPT now uses gpt-4-turbo.
tikkun··on Ask HN: How do you start a research based company?
Link?
tikkun··on Ask HN: Buying Yourself a Job
> I've considered buying a coffee shop or maker space or something and just hanging out with my customers for a living, which sounds nice in theory, but a lot of people will say dont do this because you're just buying yourself a job, and I can see that

It sounds to me like you would prefer to go ahead with the coffee shop / maker space thing, but you kinda feel like you "shouldn't" do it because people have told you not to - but you still kinda want to.

My general experience is that when I have a peaceful preference towards something, especially when it's something other people have told me I shouldn't, and especially when it's been true for months (not just a fleeting one week idea), then when I end up doing it, I feel particularly glad that I did. And when I don't follow my peaceful preference / my gut and go with it, I tend to lastingly regret not doing so. (So, perhaps consider some meditation i.e. sitting with eyes closed and kinda talking to yourself to get clearer on your intuitions - what do you feel you "should"/"shouldn't" do, what do you have impulses to do/avoid, and what do you overall peacefully prefer?)

My suggestion would be to trust your gut! Consider trying the coffee shop / maker space thing, in a small way. But if you still wanna just go big and buy one, then do that.

But you could try it out by renting a temporary space for a week or a month, and creating a pop up cafe or maker space. Could work especially well during holidays or alongside any special events.

tikkun··on Tell HN: Submit comments to IRS re tax treatment of software dev expenses
There are 43 comments so far. I was expecting thousands of comments.

Comments can be with your name, company name, or anonymous.

There's no required info you need to give other than the comment, no account required. Submitting a comment takes 'time to write comment + 5 seconds' - it was very easy.

The comments have been open for 58 days, and they close in 20 days.

Side note: Which of our reps should we call about this, how do we find which rep applies to us, and how can we contact them and what to say? It'd be nice if someone made something like Resistbot for this.

(Though, after checking out the updated version of Resistbot, it seems like Resistbot will in fact work for this. https://resist.bot)

tikkun··on Ask HN: Which DNS service do you use for your computer's network settings?
Added to OP: default DNS -> my ISP's default DNS.
tikkun··on 1Password detects "suspicious activity" in its internal Okta account
What's the worst case? My understanding is that everything is encrypted at rest and in transit, therefore even if there's a breach, user passwords would be secure.
← PreviousPage 5 of 25Next →