We need to summarize 4 most prevalent themes of opinions expressed in this Hacker News discussion. The discussion is long, covering topics: AI impact on voice actors, unionization, self-interest vs collective, automation history, AI tools, etc. Need to identify top 4 themes.
We need to read through discussion and find recurring patterns. Let's parse.
Comments:
petcat: wonders if voice real or AI, curious how industries dealing.
newyankee: stages of grief; gatekeepers hold onto hallucination errors.
lambda: mentions screen/voice actors unionized, striking; devs not unionized, no collective bargaining.
r_lee: proud to work so they can be replaced (ironic). quaintdev: "It won't happen to me". georgemcbay: expands on blind spot, organizations automate away labor, etc.
embedding-shape: says many spent career trying to automate themselves away.
12ags-18276: mentions nerd heroism lore; some dislike automation; enjoy constructing virtual worlds.
genericone: HN people hate doing 'the thing' and build systems to avoid doing it.
12ags-18276 again: many who promoted DEI fired later; "useful idiots".
zamadatix: supports unions for common good, doesn't want to bar calculators; wants spoils distribution; AI gave more work currently.
oblio: agrees unions can help distribute spoils.
zamadatix: clarifies should use unions etc not to hide from replacement.
BeetleB: computing history about replacing humans; if proud pre-LLMs, consistent; hypocritical to change now.
qwytw: if 10x engineer become 100x; else hold on or new career.
GeorgeWBasic: career turning into shit because management tries to turn them into reverse centaurs.
catigula: mentions racial diversity detrimental to unionization; Amazon used to bust Somali unions; link.
12ahsgf: scientific paper.
Lagoon3384_SWE: Sweden office worker union; cannot prevent offshoring or AI.
arjie: can't be self-interested; doesn't want to stop others using AI; adapt.
koe123: unionizing protects vulnerable; not advocating for rights weakens others.
smallerize: residential height limits; not anti-progress.
BeetleB: height limiting is anti-progress.
smallerize: wanting something good for yourself and peers isn't anti-progress if benefits accrue only to few.
BeetleB: wanting something good for yourself and peers is anti-progress if makes worse for others.
lambda: unions wouldn't necessarily negotiate no AI; SAG-AFRA deals require consent and negotiation; union gives better seat at table.
iAMkenough: quoting earlier comment; suggests self-interest leads to wage disparities.
brendoelfrendo: unionizing is collective bargaining for group benefit.
jackdoe: group wins, rest loses; localized gains, distributed loss.
computerliker: as art enjoyer, less media by computer is loss; computer never sad/horny/taste; artists should be paid living wages; thanks unions.
binary132: no moral imperative to help world reach optimum; competing interests.
sib: Carpenters' Union pushing against initiative to build more housing because reduces union labor requirements; unions in-group vs out-group.
magicalist: rebuttal; says unions are in-group vs out-group phenomenon; other side? forming corps etc.
baron816: examples of public sector strikes; better working conditions via competition; unions deter new employers.
mattkevan: maybe if employers offered better conditions people wouldn't strike.
mchusma: Baron's examples are public sector unions; employer is citizens; voluntary unions okay; could be support structures not mafia.
lambdas: blame employer who saw writing on wall.
baron816: unions overreach and cripple industries.
itishappy: self-interested unions stay away from my interests.
csallen: not against self-interested nature of unionizing; against trying to stop/hinder new tech because personally threatening.
jplusequalt: caring about ability to pay bills is self-interested like not driving into traffic.
popalchemist: collective bargaining for little guys to stand up to 1%; fight abuses.
timr: difference between stealing voice/identity (unfair) and preventing automation (sour grapes); can't get incensed over robot writing code.
boredatoms: No union offered.
lambda: problem is someone has to start union; takes work; many software engineers think too smart/professional to need union; management may block.
vlyan: quotes union demanded protections; California legislation AB 2602 and AB 1836; says law not win because those refusing consent will be passed over; no law forces humans over machines; industry may move elsewhere.
isoprophlex: developers remain ionized? (unclear)
dzonga: developers need co-ops French style.
ameliaquining: employees front capital?
epolanski: developers enjoyed scarcity; could barter themselves; majority non-unionized; can't stop tech changes by unionizing.
ameliaquining: advocates can't explain how unionized engineer works.
earthnail: work without personal brand dead; arbitrary voice acting dead; brutal.
moscoe: perfect application of AI; some jobs should be automated.
tiagod: prefer human voices; not alone; sentiment will grow.
chung8123: as long as enough people there will be market; many prefer human but don't want human cost.
oblio: which jobs should not be automated?
porridgeraisin: many jobs will get automated like past; voice actor not eternal; similar to draftsman; next gen must upskill to operate/build tools; current gen will disperse.
For software engineers, this primitive is software so easier upskilling path to build it.
bubblemoth: automation timeline faster; fewer other fields to go to; white collar automation; where will people go? uncomfortable period.
porridgeraisin: transition slower than portrayed; slow transition helps answer where will they go; ultimately everything serves social/political/personal human incentives; discounted in takeoff predictions.
space_fountain: sad about little creative control; tools great for slop; wants turning photo into rigged 3d model etc.
bonoboTP: text not only input; can provide 3d block outs etc.
NooneAtAll3: uncanny valley from youtube.
thangalin: audio books soon; locally hosted app to narrate sci-fi novel; cost vs local solution.
dormento: imagines Silly Tavern with dialog tagging.
ianmarcinkowski: would pay actors for license to voice.
thangalin: from reddit rates.
ChickeNES: no unified beloved actors.
unified101: celebrity old world concept.
dominotw: AI effects software dev because AI has compilers, tests, CI, bought tons of data in mercor; always sucks at else.
037: prompt engineering tip for Google employees.
fg137: Google ignoring Firefox.
fl4regun: not doing good job.
Firefox makes up about 90% of Mozilla’s revenue... link.
jorl17: Google wants Firefox survive for anti-competitive reasons.
sva_: Hot take: Google keeps Mozilla/Firefox alive through default search engine placement so they don't get designated as monopoly.
BbzzbB: lukewarm.
Vinnl: same for Apple/Safari.
oblio: Billion with a B.
sva_: Apple forcing users to use Safari doesn't disprove.
static_motion: factual.
WarmWash: Google internally fractured; only chrome people care.
oblio: if anything FF left out because usage low.
mattlondon: heard Firefox banned internally at Google.
tvst: not true up to 2018.
mattlondon: more recent.
GaggiX: Firefox struggle with demo pages of text-to-video models.
ygouzerh: webpage horrible.
Gecko4072: Seedance good due TikTok and YouTube; wonder proportion of private video; data labeling; next evolution just access; same for LLMs; would people sell data? messed up way to make yourself obsolete.
guilhermeasper: Google does anything except launch new Gemini Pro.
AISnakeOil: why need? instant models more important for search; Pro models for coding agent work; not money.
sva_: also problem of given away pro subscription to students; tightening down.
jimmoores: stole my money by cutting me off from Gemini CLI; no explanation.
SimianSci: capital infusion frontier labs huge; may not recoup without unrealistic things.
jimmoores: Chinese labs can compete on shoestring; Google should compete.
bitexploder: AI/LLM more than agentic coding; HN over-indexed.
SimianSci: not shoestring; state lenders cheap credit.
msabalau: could make larger models slower cadence.
onlyrealcuzzo: Google paid for 3.5 Pro training; never gave answer; they spent money.
bitexploder: works at Google; zero internal knowledge; opinion own; not worth fighting month to month; Gemini Pro 3.1 released Feb; feels forever; but 2-3 yr horizon similar; Flash 3.5 to 3.7 etc.
mh-: CTO of GCP shop; if Google still wouldn't have competitive offering to agentic coding by now would not have believed.
WarmWash: 3.5 was almost certainly a 3.1 post-train; small investment; started pretraining Gemini 4.
guilhermeasper: agree race doesn't make sense; rumors Google produced newer Pro not released.
SimianSci: cost of model includes training (Capex) low OpEx after; may wish to slow cadence, focus elsewhere.
gbriel: Google doesn't have good coding model; need good coding model for competitive advantage.
dbbk: competitive advantage why? will make more money? Already have Google Cloud.
jnwatson: 3.7 flash ain't bad; fits second tier.
dbbk: all Google has to do is build model works good enough for Gemini app and Spark; they have it.
mianos: have pro subscription; think they have given up; because test new models against frontier they are bad; pull back; innovate elsewhere less competition.
epolanski: you live in HN/tech bubble; real world Google/Microsoft dominate enterprise; non-tech office workers writing Gemini gems or Copilot; they already have Google/MS; procurement etc sorted.
They don't need best coding model; need better/cheaper while serving existing base.
Microsoft can provide any open/closed AI with services, focus on harness.
vismit2000: new Gemini Pro like GTA-6.
simonw: Interesting OpenAI abandoned Sora; Google investing heavily in video generation; maybe for world models.
kfarr: Google has huge built-in training corpus with YouTube, gphotos, geospatial.
OtherShrezzing: 10% revenue from YouTube; need to own tech that might disrupt.
bahmboo: paid API access only; here to make money not IPO.
msabalau: Google committed to multimodal; Veo and omni better than Sora; YouTube huge; microdramas category.
pwython: Sora 2 better than Veo quality but limited use case for businesses.
mattlondon: YouTube and video ads; previously needed actors etc; now video ads just a prompt.
WarmWash: Google owns 15% of Anthropic; Claude trains on TPUs; Google cloud backlogged with demand from OAI and Anthropic; selling shovels etc.
simonw: convincing answer; many advertisers need video ads.
chermi: Google's main edge multimodal; still best overall? would try be best at something; not excited about Sora etc; love read-side.
kranke155: Sora social network type; Google sells models PAYG; Nano Banana changed advertising 2D mockups and Photoshop; GPT image 2 also PAYG.
Work in advertising; spent lot using models; amount and rapidity changed everything.
tiahura: wish models supported 3d.
CmonGoogle: AI brand fragmentation at Google not problem because pretending: SOTA models more effective; OpenAI/Anthropic lead; Google's moat is search bread/butter; all 3 false; Google connected standard LLMs to externally valuable live service; as more companies realize value is service not AI capability, products/humans become important again; Google should be Google again; Gemini off to side; Omni confuses; resolve AI mode; rename Gemma; consolidate brand.
bahmboo: Google is search; implicit to that is "find"; AI integration into search hit stride; they have search box hard wired into people; iterating crafting AI into experience; they failed initially.
CmonGoogle: YMMV; gets acronyms wrong; assumes mean something else; search engine working then LLM ruining.
Other UIs besides chat.
polytely: uncanny valley from videos featuring humans disgusts.
aloknnikhil: eyes robotic/lifeless; sound mixing off; voice layered on background not blended.
kzrdude: can't see difference; skin crawl expecting something off.
bahmboo: knew upfront generated; looked at lips for flaws; fish swimming can't tell real vs generated even deep inspection.
overflowy: using generated videos in professional setting cannot ever use business.
visarga: Only AI code holy, video sinful.
rcr-anti: easy demos but struggle practical application; work/enjoyment from this? Ads/media pre-production plausible; fails 'how can this enrich life' in way most other AI tools don't; maybe if only interest is other meaning of enrich that might flow from ads and numbing rivers of slop.
Why do we look at art/watch videos? replicable as function of text, other media, 3-30 cents compute per sec? functionalist about these; at some point won't tell difference; but until then might say 'death of the author', category error.
I work with artists using video/image generation models; they interested in faster iteration and controlling intermediate steps (graphs labyrinthine).
msabalau: most people using AI for creative projects spend time mastering tools, adapt into creative processes; AI lowers cost indie productions, allows broader range stories; even successful film makers need to bow/scrape to get funded; democratizing visual media good; even if personally no more likely to do this than pick up Photoshop or record podcast.
I enjoy making short films with AI; when my latest short screens at festival Oct alongside traditional and AI films, hopefully audience will like it.
Quick "one shot" video generation might be slop to you or I; but if someone wants to send as birthday greeting to aunt and they both enjoy it, what business is it of ours?
visarga: cheaper to produce, more daring it can be; good thing.
jiggawatts: kids love image/video generation; former cheap enough to do just because fun; have young boy; builds LEGO scene; take reference pictures; make into "real" movie scene, cartoon etc; he loves it, motivates to build bigger; if really special might fork $5 to use Omni to turn LEGO creation into 10-sec video instead of still image; blows his mind.
PS: cheap/free phone apps make stop-motion trivial; already made couple videos of toys moving that way.
thisisauserid: Google's main revenue is advertising not enriching lives; not charity.
visarga: as in what work/enjoyment someone gets; why look at art/watch videos? I like to generate songs from obscure poems.
Nihilartikel: got mildly excited with last Omni release but can't do practical thing I want - sync generated video to pre-existing audio; happily using Minimax H3 locally on 12Gb 4070RTX to finish lip syncing to recorded dialog on abandoned 20 yr Flash animation hobby projects.
msabalau: yeah that'd be great feature.
jarjoura: pretty sure heard PM in podcast say intentionally not building support for it out of concerns enabling deep-fakes; agree though; cost for using AI video models too high for anyone not building anything serious; too restricted for using professionally; prompting them with text to get something generated cute but then you just create slop everyone hates, devaluing power.
doctorpangloss: Deepfakes are kind of the point though.
SpyCoder77: making up numbers now huh.
shreya1999: raw generation quality table stakes; controllability more important battleground.
chermi: Google we ain't falling for it. #neverforget3.5prowithinamonth
cube00: Draft videos more efficiently in 360p; while sounds great you quickly disappointed after run same prompt at standard resolution only to get different result because non deterministic.
kridsdale1: can you hold seed constant?
cube00: no option in Web UI; changelog suggests using new upscaler but had nothing but disappointment from upscalers.
Upscale when ready: Users on paid tiers can seamlessly upgrade favorite clips using our new 360p to 720p upscaler.
peter_d_sherman: >"Draft videos more efficiently in 360p
Generate lightweight previews in 360p resolution up to 60% faster
and at a third of the cost compared to Omni 1.1’s standard 720p resolution. This is helpful for rapid prototyping, storyboard iteration, and quick rendering in developer platforms."
This is great idea to have low-resolution mode for additional speed to create previews, do test runs, create rapid prototypes, etc.
My curiosity: what's absolute usable minimum that this could be? Would/could 240p work? 144p? lower? then could those images be upscaled quickly (and result still usable?) with faster image upscaling-only neural network?
Reason: knowing lower numbers could be important for additional cost/time savings and/or running derived LLM's on local resource-constrained hardware...
Anyway, great post, great idea, welcome Gemini Omni 1.1 Flash to ever-expanding list of LLM/AI's!
Now we need to identify 4 most prevalent themes. Let's tally topics:
- Impact of AI on creative professions (voice actors, artists) and concerns about replacement, authenticity, unions protecting them. (petcat, lambda, georgemcbay, embedding-shape, BeetleB, qwytw, GeorgeWBasic, cat